chnjinlei/hive
1
0
Fork 0
Code Issues Pull Requests Actions 39 Packages Projects Releases Wiki Activity

fix: resolve all ruff lint and format errors across codebase (#7058)

Browse Source 
- Auto-fixed 70 lint errors (import sorting, aliased errors, datetime.UTC)
- Fixed 85 remaining errors manually:
  - E501: wrapped long lines in queen_profiles, catalog, routes_credentials
  - F821: added missing TYPE_CHECKING imports for AgentHost, ToolRegistry,
    HookContext, HookResult; added runtime imports where needed
  - F811: removed duplicate method definitions in queen_lifecycle_tools
  - F841/B007: removed unused variables in discovery.py
  - W291: removed trailing whitespace in queen nodes
  - E402: moved import to top of queen_memory_v2.py
  - Fixed AgentRuntime -> AgentHost in example template type annotations
- Reformatted 343 files with ruff format
This commit is contained in:
Hundao
2026-04-16 19:30:01 +08:00
committed by GitHub
parent 4fdbc438f9
commit 589c5b06fe
354 changed files with 1955 additions and 5340 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.claude/skills/browser-edge-cases/scripts/test_02_twitter_scroll.py
+3 -6
View File
@@ -57,8 +57,7 @@ async def test_twitter_lazy_scroll():
# Count initial tweets
initial_count = await bridge.evaluate(
tab_id,
"(function() { return document.querySelectorAll("
"'[data-testid=\"tweet\"]').length; })()",
"(function() { return document.querySelectorAll('[data-testid=\"tweet\"]').length; })()",
)
print(f"Initial tweet count: {initial_count.get('result', 0)}")
@@ -78,8 +77,7 @@ async def test_twitter_lazy_scroll():
# Count tweets after scroll
count_result = await bridge.evaluate(
tab_id,
"(function() { return document.querySelectorAll("
"'[data-testid=\"tweet\"]').length; })()",
"(function() { return document.querySelectorAll('[data-testid=\"tweet\"]').length; })()",
)
count = count_result.get("result", 0)
print(f" Tweet count after scroll: {count}")
@@ -87,8 +85,7 @@ async def test_twitter_lazy_scroll():
# Final count
final_count = await bridge.evaluate(
tab_id,
"(function() { return document.querySelectorAll("
"'[data-testid=\"tweet\"]').length; })()",
"(function() { return document.querySelectorAll('[data-testid=\"tweet\"]').length; })()",
)
final = final_count.get("result", 0)
initial = initial_count.get("result", 0)
.claude/skills/browser-edge-cases/scripts/test_06_shadow_dom.py
+1 -3
View File
@@ -130,9 +130,7 @@ async def test_shadow_dom():
print(f"JS click result: {click_result.get('result', {})}")
# Verify click was registered
count_result = await bridge.evaluate(
tab_id, "(function() { return window.shadowClickCount || 0; })()"
)
count_result = await bridge.evaluate(tab_id, "(function() { return window.shadowClickCount || 0; })()")
count = count_result.get("result") or 0
print(f"Shadow click count: {count}")
.claude/skills/browser-edge-cases/scripts/test_08_autocomplete.py
+2 -5
View File
@@ -200,9 +200,7 @@ async def test_autocomplete():
print(f"Value after fast typing: '{fast_value}'")
# Check events
events_result = await bridge.evaluate(
tab_id, "(function() { return window.inputEvents; })()"
)
events_result = await bridge.evaluate(tab_id, "(function() { return window.inputEvents; })()")
print(f"Events logged: {events_result.get('result', [])}")
# Test 2: Slow typing (with delay) - should work
@@ -220,8 +218,7 @@ async def test_autocomplete():
# Check if dropdown appeared
dropdown_result = await bridge.evaluate(
tab_id,
"(function() { return document.querySelectorAll("
"'.autocomplete-items div').length; })()",
"(function() { return document.querySelectorAll('.autocomplete-items div').length; })()",
)
dropdown_count = dropdown_result.get("result", 0)
print(f"Dropdown items: {dropdown_count}")
.claude/skills/browser-edge-cases/scripts/test_10_huge_dom.py
+3 -9
View File
@@ -87,9 +87,7 @@ async def test_huge_dom():
await bridge.navigate(tab_id, data_url, wait_until="load")
# Count elements
count_result = await bridge.evaluate(
tab_id, "(function() { return document.querySelectorAll('*').length; })()"
)
count_result = await bridge.evaluate(tab_id, "(function() { return document.querySelectorAll('*').length; })()")
elem_count = count_result.get("result", 0)
print(f"DOM elements: {elem_count}")
@@ -122,14 +120,10 @@ async def test_huge_dom():
# Test 3: Real LinkedIn
print("\n--- Test 3: Real LinkedIn Feed ---")
await bridge.navigate(
tab_id, "https://www.linkedin.com/feed", wait_until="load", timeout_ms=30000
)
await bridge.navigate(tab_id, "https://www.linkedin.com/feed", wait_until="load", timeout_ms=30000)
await asyncio.sleep(2)
count_result = await bridge.evaluate(
tab_id, "(function() { return document.querySelectorAll('*').length; })()"
)
count_result = await bridge.evaluate(tab_id, "(function() { return document.querySelectorAll('*').length; })()")
elem_count = count_result.get("result", 0)
print(f"LinkedIn DOM elements: {elem_count}")
.claude/skills/browser-edge-cases/scripts/test_15_screenshot.py
+2 -7
View File
@@ -136,10 +136,7 @@ async def test_selector_screenshot(bridge: BeelineBridge, tab_id: int, data_url:
print(" ⚠ WARNING: Selector screenshot not smaller (may be full page)")
return False
else:
print(
" ⚠ NOT IMPLEMENTED: selector param ignored"
f" (returns full page) - error={result.get('error')}"
)
print(f" ⚠ NOT IMPLEMENTED: selector param ignored (returns full page) - error={result.get('error')}")
print(" NOTE: selector parameter exists in signature but is not used in implementation")
return False
@@ -181,9 +178,7 @@ async def test_screenshot_timeout(bridge: BeelineBridge, tab_id: int, data_url:
print(f" ⚠ Fast enough to beat timeout: {err!r} in {elapsed:.3f}s")
return True # Not a failure, just fast
else:
print(
f" ⚠ Screenshot completed before timeout ({elapsed:.3f}s) - too fast to test timeout"
)
print(f" ⚠ Screenshot completed before timeout ({elapsed:.3f}s) - too fast to test timeout")
return True # Still ok, just very fast
.claude/skills/browser-edge-cases/scripts/test_case.py
+2 -8
View File
@@ -137,14 +137,8 @@ async def test_problematic_site(bridge: BeelineBridge, tab_id: int) -> dict:
changed = False
for key in after_data:
if key in before_data:
b_val = (
before_data[key].get("scrollTop", 0)
if isinstance(before_data[key], dict)
else 0
)
a_val = (
after_data[key].get("scrollTop", 0) if isinstance(after_data[key], dict) else 0
)
b_val = before_data[key].get("scrollTop", 0) if isinstance(before_data[key], dict) else 0
a_val = after_data[key].get("scrollTop", 0) if isinstance(after_data[key], dict) else 0
if a_val != b_val:
print(f" ✓ CHANGE DETECTED: {key} scrolled from {b_val} to {a_val}")
changed = True
core/antigravity_auth.py
+7 -21
View File
@@ -52,9 +52,7 @@ _DEFAULT_REDIRECT_PORT = 51121
# This project reverse-engineered and published the public OAuth credentials
# for Google's Antigravity/Cloud Code Assist API.
# Source: https://github.com/NoeFabris/opencode-antigravity-auth
_CREDENTIALS_URL = (
"https://raw.githubusercontent.com/NoeFabris/opencode-antigravity-auth/dev/src/constants.ts"
)
_CREDENTIALS_URL = "https://raw.githubusercontent.com/NoeFabris/opencode-antigravity-auth/dev/src/constants.ts"
# Cached credentials fetched from public source
_cached_client_id: str | None = None
@@ -68,9 +66,7 @@ def _fetch_credentials_from_public_source() -> tuple[str | None, str | None]:
return _cached_client_id, _cached_client_secret
try:
req = urllib.request.Request(
_CREDENTIALS_URL, headers={"User-Agent": "Hive-Antigravity-Auth/1.0"}
)
req = urllib.request.Request(_CREDENTIALS_URL, headers={"User-Agent": "Hive-Antigravity-Auth/1.0"})
with urllib.request.urlopen(req, timeout=10) as resp:
content = resp.read().decode("utf-8")
import re
@@ -168,10 +164,7 @@ class OAuthCallbackHandler(BaseHTTPRequestHandler):
if "code" in query and "state" in query:
OAuthCallbackHandler.auth_code = query["code"][0]
OAuthCallbackHandler.state = query["state"][0]
self._send_response(
"Authentication successful! You can close this window "
"and return to the terminal."
)
self._send_response("Authentication successful! You can close this window and return to the terminal.")
return
self._send_response("Waiting for authentication...")
@@ -296,8 +289,7 @@ def validate_credentials(access_token: str, project_id: str = _DEFAULT_PROJECT_I
"Authorization": f"Bearer {access_token}",
"Content-Type": "application/json",
"User-Agent": (
"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) "
"AppleWebKit/537.36 (KHTML, like Gecko) Antigravity/1.18.3"
"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Antigravity/1.18.3"
),
"X-Goog-Api-Client": "google-cloud-sdk vscode_cloudshelleditor/0.1",
}
@@ -316,9 +308,7 @@ def validate_credentials(access_token: str, project_id: str = _DEFAULT_PROJECT_I
return False
def refresh_access_token(
refresh_token: str, client_id: str, client_secret: str | None
) -> dict | None:
def refresh_access_token(refresh_token: str, client_id: str, client_secret: str | None) -> dict | None:
"""Refresh the access token using the refresh token."""
data = {
"grant_type": "refresh_token",
@@ -361,9 +351,7 @@ def cmd_account_add(args: argparse.Namespace) -> int:
access_token = account.get("access")
refresh_token_str = account.get("refresh", "")
refresh_token = refresh_token_str.split("|")[0] if refresh_token_str else None
project_id = (
refresh_token_str.split("|")[1] if "|" in refresh_token_str else _DEFAULT_PROJECT_ID
)
project_id = refresh_token_str.split("|")[1] if "|" in refresh_token_str else _DEFAULT_PROJECT_ID
email = account.get("email", "unknown")
expires_ms = account.get("expires", 0)
expires_at = expires_ms / 1000.0 if expires_ms else 0.0
@@ -390,9 +378,7 @@ def cmd_account_add(args: argparse.Namespace) -> int:
# Update the account
account["access"] = new_access
account["expires"] = int((time.time() + expires_in) * 1000)
accounts_data["last_refresh"] = time.strftime(
"%Y-%m-%dT%H:%M:%SZ", time.gmtime()
)
accounts_data["last_refresh"] = time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
save_accounts(accounts_data)
# Validate the refreshed token
core/framework/agent_loop/agent_loop.py
+51 -159
View File
@@ -126,9 +126,7 @@ _STRIP_RE = re.compile(
# The value cannot contain `<` or `\n` — those terminate the label.
# Trailing whitespace (including the terminating newline) is consumed
# so the visible text that follows starts cleanly.
_LABEL_STRIP_RE = re.compile(
r"<(?:" + "|".join(_INTERNAL_TAGS) + r")>[^<\n]*\s*"
)
_LABEL_STRIP_RE = re.compile(r"<(?:" + "|".join(_INTERNAL_TAGS) + r")>[^<\n]*\s*")
# Matches a trailing `<` that could be the start of an internal tag.
# We build a pattern that matches `<` followed by any prefix of any
@@ -138,9 +136,7 @@ for _tag in _INTERNAL_TAGS:
for _i in range(1, len(_tag) + 1):
_PARTIAL_PREFIXES.add(_tag[:_i])
_PARTIAL_OPEN_RE = re.compile(
r"<(?:"
+ "|".join(re.escape(p) for p in sorted(_PARTIAL_PREFIXES, key=len, reverse=True))
+ r")$"
r"<(?:" + "|".join(re.escape(p) for p in sorted(_PARTIAL_PREFIXES, key=len, reverse=True)) + r")$"
)
_GENERIC_TAG_RE = re.compile(r"</?[a-zA-Z_][\w-]*\s*/?>")
@@ -351,9 +347,7 @@ class AgentLoop(AgentProtocol):
self._config = config or LoopConfig()
self._tool_executor = tool_executor
self._conversation_store = conversation_store
self._injection_queue: asyncio.Queue[tuple[str, bool, list[dict[str, Any]] | None]] = (
asyncio.Queue()
)
self._injection_queue: asyncio.Queue[tuple[str, bool, list[dict[str, Any]] | None]] = asyncio.Queue()
self._trigger_queue: asyncio.Queue[TriggerEvent] = asyncio.Queue()
# Queen input blocking state
self._input_ready = asyncio.Event()
@@ -510,9 +504,7 @@ class AgentLoop(AgentProtocol):
output_tokens=0,
latency_ms=0,
)
return self._finalize_result(
AgentResult(success=False, error=error_msg), "guard_failure"
)
return self._finalize_result(AgentResult(success=False, error=error_msg), "guard_failure")
# 2. Restore or create new conversation + accumulator
restored = await self._restore(ctx)
@@ -571,11 +563,7 @@ class AgentLoop(AgentProtocol):
if ctx.default_skill_batch_nudge:
from framework.skills.defaults import is_batch_scenario as _is_batch
_input_text = (
(ctx.goal_context or "")
+ " "
+ " ".join(str(v) for v in ctx.input_data.values() if v)
)
_input_text = (ctx.goal_context or "") + " " + " ".join(str(v) for v in ctx.input_data.values() if v)
if _is_batch(_input_text):
system_prompt = f"{system_prompt}\n\n{ctx.default_skill_batch_nudge}"
logger.info("[%s] DS-12: batch scenario detected, nudge injected", node_id)
@@ -587,9 +575,7 @@ class AgentLoop(AgentProtocol):
store=self._conversation_store,
run_id=ctx.effective_run_id,
compaction_buffer_tokens=self._config.compaction_buffer_tokens,
compaction_warning_buffer_tokens=(
self._config.compaction_warning_buffer_tokens
),
compaction_warning_buffer_tokens=(self._config.compaction_warning_buffer_tokens),
)
accumulator = OutputAccumulator(
store=self._conversation_store,
@@ -690,9 +676,7 @@ class AgentLoop(AgentProtocol):
node_id,
iteration,
)
await self._publish_loop_completed(
stream_id, node_id, iteration, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration, execution_id)
return AgentResult(
success=True,
output=accumulator.to_dict(),
@@ -773,9 +757,7 @@ class AgentLoop(AgentProtocol):
prompt=str(pending_input_state.get("prompt", "")),
options=pending_input_state.get("options"),
questions=pending_input_state.get("questions"),
emit_client_request=bool(
pending_input_state.get("emit_client_request", True)
),
emit_client_request=bool(pending_input_state.get("emit_client_request", True)),
)
logger.info(
"[%s] iter=%d: restored wait unblocked, got_input=%s",
@@ -784,9 +766,7 @@ class AgentLoop(AgentProtocol):
got_input,
)
if not got_input:
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
return AgentResult(
success=True,
@@ -797,8 +777,7 @@ class AgentLoop(AgentProtocol):
)
if self._injection_queue.empty() and self._trigger_queue.empty():
logger.info(
"[%s] iter=%d: pending-input wait woke"
" without queued input; re-waiting",
"[%s] iter=%d: pending-input wait woke without queued input; re-waiting",
node_id,
iteration,
)
@@ -863,9 +842,7 @@ class AgentLoop(AgentProtocol):
iteration,
len(conversation.messages),
)
logger.debug(
"[AgentLoop.execute] iteration=%d: entering _run_single_turn loop", iteration
)
logger.debug("[AgentLoop.execute] iteration=%d: entering _run_single_turn loop", iteration)
_stream_retry_count = 0
_capacity_retry_started_at: float | None = None
_capacity_retry_attempt = 0
@@ -892,9 +869,7 @@ class AgentLoop(AgentProtocol):
request_system_prompt,
request_messages,
_,
) = await self._run_single_turn(
ctx, conversation, tools, iteration, accumulator
)
) = await self._run_single_turn(ctx, conversation, tools, iteration, accumulator)
logger.debug(
"[AgentLoop.execute] iteration=%d: _run_single_turn completed successfully",
iteration,
@@ -910,10 +885,7 @@ class AgentLoop(AgentProtocol):
len(real_tool_results),
outputs_set or "[]",
turn_tokens,
{
k: ("set" if v is not None else "None")
for k, v in accumulator.to_dict().items()
},
{k: ("set" if v is not None else "None") for k, v in accumulator.to_dict().items()},
)
total_input_tokens += turn_tokens.get("input", 0)
total_output_tokens += turn_tokens.get("output", 0)
@@ -982,10 +954,7 @@ class AgentLoop(AgentProtocol):
# still publishes a retry event so the UI can see us
# waiting (the "heartbeat" — no silent stalls).
self._bump("llm_turn_exception")
if (
self._is_capacity_error(e)
and self._config.capacity_retry_max_seconds > 0
):
if self._is_capacity_error(e) and self._config.capacity_retry_max_seconds > 0:
self._bump("capacity_error")
now = time.monotonic()
if _capacity_retry_started_at is None:
@@ -994,8 +963,7 @@ class AgentLoop(AgentProtocol):
if elapsed < self._config.capacity_retry_max_seconds:
_capacity_retry_attempt += 1
delay = min(
self._config.stream_retry_backoff_base
* (2 ** min(_capacity_retry_attempt - 1, 6)),
self._config.stream_retry_backoff_base * (2 ** min(_capacity_retry_attempt - 1, 6)),
self._config.capacity_retry_max_delay,
)
logger.warning(
@@ -1023,15 +991,11 @@ class AgentLoop(AgentProtocol):
continue # retry same iteration
# Retry transient errors with exponential backoff
if (
self._is_transient_error(e)
and _stream_retry_count < self._config.max_stream_retries
):
if self._is_transient_error(e) and _stream_retry_count < self._config.max_stream_retries:
self._bump("llm_transient_retry")
_stream_retry_count += 1
delay = min(
self._config.stream_retry_backoff_base
* (2 ** (_stream_retry_count - 1)),
self._config.stream_retry_backoff_base * (2 ** (_stream_retry_count - 1)),
self._config.stream_retry_max_delay,
)
logger.warning(
@@ -1079,8 +1043,7 @@ class AgentLoop(AgentProtocol):
if ctx.supports_direct_user_io:
error_msg = f"LLM call failed: {e}"
_guardrail_phrase = (
"no endpoints available matching your guardrail restrictions "
"and data policy"
"no endpoints available matching your guardrail restrictions and data policy"
)
if _guardrail_phrase in str(e).lower():
error_msg += (
@@ -1219,9 +1182,7 @@ class AgentLoop(AgentProtocol):
node_id,
iteration,
)
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
return AgentResult(
success=True,
@@ -1442,10 +1403,7 @@ class AgentLoop(AgentProtocol):
_has_tools_no_text = bool(real_tool_results) and not assistant_text
if _has_tools_no_text:
_silent_tool_streak += 1
if (
_silent_tool_streak > 0
and _silent_tool_streak % self._config.silent_tool_streak_threshold == 0
):
if _silent_tool_streak > 0 and _silent_tool_streak % self._config.silent_tool_streak_threshold == 0:
nudge = (
"[SYSTEM] You have been calling tools for "
f"{_silent_tool_streak} consecutive turns without "
@@ -1488,10 +1446,7 @@ class AgentLoop(AgentProtocol):
and self._event_bus is not None
)
_worker_no_tool_turn = (
not real_tool_results
and not outputs_set
and not queen_input_requested
and not user_input_requested
not real_tool_results and not outputs_set and not queen_input_requested and not user_input_requested
)
if _is_worker and _worker_no_tool_turn:
_worker_text_only_streak += 1
@@ -1599,9 +1554,7 @@ class AgentLoop(AgentProtocol):
step_index=iteration,
verdict="CONTINUE",
verdict_feedback=(
"Auto-block grace"
f" ({_cf_text_only_streak}"
f"/{self._config.cf_grace_turns})"
f"Auto-block grace ({_cf_text_only_streak}/{self._config.cf_grace_turns})"
),
tool_calls=logged_tool_calls,
llm_text=assistant_text,
@@ -1614,9 +1567,7 @@ class AgentLoop(AgentProtocol):
# through to judge
if self._shutdown:
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
_continue_count += 1
if ctx.runtime_logger:
@@ -1702,9 +1653,7 @@ class AgentLoop(AgentProtocol):
)
logger.info("[%s] iter=%d: unblocked, got_input=%s", node_id, iteration, got_input)
if not got_input:
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
_continue_count += 1
if ctx.runtime_logger:
@@ -1800,9 +1749,7 @@ class AgentLoop(AgentProtocol):
# until the queen injects guidance.
if queen_input_requested:
if self._shutdown:
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
_continue_count += 1
self._log_skip_judge(
@@ -1870,15 +1817,11 @@ class AgentLoop(AgentProtocol):
stream_id=stream_id,
node_id=node_id,
reason="Blocked waiting for queen guidance - no input received",
context=(
"Worker escalated but received no queen guidance before shutdown"
),
context=("Worker escalated but received no queen guidance before shutdown"),
execution_id=execution_id,
request_id=uuid.uuid4().hex,
)
await self._publish_loop_completed(
stream_id, node_id, iteration + 1, execution_id
)
await self._publish_loop_completed(stream_id, node_id, iteration + 1, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
_continue_count += 1
self._log_skip_judge(
@@ -2143,9 +2086,7 @@ class AgentLoop(AgentProtocol):
continue
# 7. Max iterations exhausted
await self._publish_loop_completed(
stream_id, node_id, self._config.max_iterations, execution_id
)
await self._publish_loop_completed(stream_id, node_id, self._config.max_iterations, execution_id)
latency_ms = int((time.time() - start_time) * 1000)
if ctx.runtime_logger:
ctx.runtime_logger.log_node_complete(
@@ -2168,9 +2109,7 @@ class AgentLoop(AgentProtocol):
return self._finalize_result(
AgentResult(
success=False,
error=(
f"Max iterations ({self._config.max_iterations}) reached without acceptance"
),
error=(f"Max iterations ({self._config.max_iterations}) reached without acceptance"),
output=accumulator.to_dict(),
tokens_used=total_input_tokens + total_output_tokens,
latency_ms=latency_ms,
@@ -2204,9 +2143,7 @@ class AgentLoop(AgentProtocol):
image_content: Optional list of OpenAI-style image blocks to attach.
"""
logger.debug(
"[AgentLoop.inject_event] content_len=%d,"
" is_client_input=%s, has_images=%s,"
" queue_size_before=%d",
"[AgentLoop.inject_event] content_len=%d, is_client_input=%s, has_images=%s, queue_size_before=%d",
len(content) if content else 0,
is_client_input,
bool(image_content),
@@ -2440,9 +2377,7 @@ class AgentLoop(AgentProtocol):
# generating. Unsafe tools (bash, edits, browser actions)
# still wait for FinishEvent so we don't race a write
# against a decision the model hasn't finished making.
_early_safe_names = {
t.name for t in tools if getattr(t, "concurrency_safe", False)
}
_early_safe_names = {t.name for t in tools if getattr(t, "concurrency_safe", False)}
_early_tasks: dict[str, asyncio.Task] = {}
async def _timed_execute(
@@ -2539,9 +2474,7 @@ class AgentLoop(AgentProtocol):
and "_raw" not in event.tool_input
and event.tool_use_id not in _tasks
):
_tasks[event.tool_use_id] = asyncio.create_task(
_exec_fn(event)
)
_tasks[event.tool_use_id] = asyncio.create_task(_exec_fn(event))
elif isinstance(event, FinishEvent):
token_counts["input"] += event.input_tokens
@@ -2558,9 +2491,7 @@ class AgentLoop(AgentProtocol):
_llm_stream_t0 = time.monotonic()
self._stream_task = asyncio.create_task(_do_stream())
logger.debug(
"[_run_single_turn] inner_turn=%d: Stream task created, waiting...", inner_turn
)
logger.debug("[_run_single_turn] inner_turn=%d: Stream task created, waiting...", inner_turn)
_inactivity_limit = self._config.llm_stream_inactivity_timeout_seconds
try:
if _inactivity_limit and _inactivity_limit > 0:
@@ -2574,9 +2505,7 @@ class AgentLoop(AgentProtocol):
# TimeoutError of its own" — wait_for conflates them.
_check_interval = min(5.0, _inactivity_limit / 2)
while True:
done, _pending = await asyncio.wait(
{self._stream_task}, timeout=_check_interval
)
done, _pending = await asyncio.wait({self._stream_task}, timeout=_check_interval)
if self._stream_task in done:
# Let any exception the task raised propagate
# naturally via the outer ``await`` below.
@@ -2607,9 +2536,7 @@ class AgentLoop(AgentProtocol):
# watchdog loop exited via ``break`` the task is done, and
# ``await`` is the cheapest way to surface its exception.
await self._stream_task
logger.debug(
"[_run_single_turn] inner_turn=%d: Stream task completed normally", inner_turn
)
logger.debug("[_run_single_turn] inner_turn=%d: Stream task completed normally", inner_turn)
except asyncio.CancelledError:
logger.debug("[_run_single_turn] inner_turn=%d: Stream task cancelled", inner_turn)
if accumulated_text:
@@ -2631,9 +2558,7 @@ class AgentLoop(AgentProtocol):
raise
raise TurnCancelled() from None
except Exception as e:
logger.exception(
"[_run_single_turn] inner_turn=%d: Stream task failed: %s", inner_turn, e
)
logger.exception("[_run_single_turn] inner_turn=%d: Stream task failed: %s", inner_turn, e)
# Don't orphan early tool tasks on a stream failure
# either - the outer retry loop will re-emit the tool
# calls on the next attempt.
@@ -2652,9 +2577,7 @@ class AgentLoop(AgentProtocol):
for _early in _early_tasks.values():
if not _early.done():
_early.cancel()
raise ConnectionError(
f"Stream failed with recoverable error: {_stream_error.error}"
)
raise ConnectionError(f"Stream failed with recoverable error: {_stream_error.error}")
final_text = accumulated_text
logger.info(
@@ -2735,19 +2658,14 @@ class AgentLoop(AgentProtocol):
# capping them strands work mid-turn and the next turn just
# re-emits the discarded calls, which is strictly worse.
if self._config.max_tool_calls_per_turn > 0:
hard_limit = int(
self._config.max_tool_calls_per_turn
* (1 + self._config.tool_call_overflow_margin)
)
hard_limit = int(self._config.max_tool_calls_per_turn * (1 + self._config.tool_call_overflow_margin))
else:
hard_limit = 0 # disabled
# Phase 1: triage — handle framework tools immediately,
# queue real tools for parallel execution.
results_by_id: dict[str, ToolResult] = {}
timing_by_id: dict[
str, dict[str, Any]
] = {} # tool_use_id -> {start_timestamp, duration_s}
timing_by_id: dict[str, dict[str, Any]] = {} # tool_use_id -> {start_timestamp, duration_s}
pending_real: list[ToolCallEvent] = []
for tc in tool_calls:
@@ -2801,9 +2719,7 @@ class AgentLoop(AgentProtocol):
sanitize_ask_user_inputs,
)
ask_user_prompt, recovered_options = sanitize_ask_user_inputs(
ask_user_prompt, raw_options
)
ask_user_prompt, recovered_options = sanitize_ask_user_inputs(ask_user_prompt, raw_options)
if recovered_options is not None and raw_options is None:
raw_options = recovered_options
# Defensive: ensure options is a list of strings.
@@ -2930,8 +2846,7 @@ class AgentLoop(AgentProtocol):
result = ToolResult(
tool_use_id=tc.tool_use_id,
content=(
"ERROR: escalate is only available to worker "
"nodes/sub-agents, not queen/judge streams."
"ERROR: escalate is only available to worker nodes/sub-agents, not queen/judge streams."
),
is_error=True,
)
@@ -2941,9 +2856,7 @@ class AgentLoop(AgentProtocol):
if self._event_bus is None:
result = ToolResult(
tool_use_id=tc.tool_use_id,
content=(
"ERROR: EventBus unavailable. Could not emit escalation request."
),
content=("ERROR: EventBus unavailable. Could not emit escalation request."),
is_error=True,
)
results_by_id[tc.tool_use_id] = result
@@ -2973,10 +2886,7 @@ class AgentLoop(AgentProtocol):
# owner (Worker instance) records the explicit report
# via ``record_explicit_report`` so Worker.run()'s
# terminal event emission picks it up.
if not (
isinstance(stream_id, str)
and stream_id.startswith("worker:")
):
if not (isinstance(stream_id, str) and stream_id.startswith("worker:")):
result = ToolResult(
tool_use_id=tc.tool_use_id,
content=(
@@ -3064,9 +2974,7 @@ class AgentLoop(AgentProtocol):
async with _sem:
return await _timed_execute(_tc)
timed_results_by_id: dict[
str, tuple[ToolResult | BaseException, str, float] | BaseException
] = {}
timed_results_by_id: dict[str, tuple[ToolResult | BaseException, str, float] | BaseException] = {}
async def _cancel_turn_with_stubs(
_pending: list[ToolCallEvent] = pending_real, # noqa: B006,B008
@@ -3108,9 +3016,7 @@ class AgentLoop(AgentProtocol):
_awaitables.append(early)
else:
_awaitables.append(_capped(tc))
self._tool_task = asyncio.ensure_future(
asyncio.gather(*_awaitables, return_exceptions=True)
)
self._tool_task = asyncio.ensure_future(asyncio.gather(*_awaitables, return_exceptions=True))
try:
parallel_timed = await self._tool_task
finally:
@@ -3196,9 +3102,7 @@ class AgentLoop(AgentProtocol):
result = _build_tool_error_result(tc, raw)
else:
result = raw
results_by_id[tc.tool_use_id] = await self._truncate_tool_result(
result, tc.tool_name
)
results_by_id[tc.tool_use_id] = await self._truncate_tool_result(result, tc.tool_name)
# Phase 3: record results into conversation in original order,
# build logged/real lists, and publish completed events.
@@ -3227,8 +3131,7 @@ class AgentLoop(AgentProtocol):
image_content = result.image_content
if image_content and ctx.llm and not supports_image_tool_results(ctx.llm.model):
logger.info(
"Stripping image_content from tool result; "
"model '%s' does not support images in tool results",
"Stripping image_content from tool result; model '%s' does not support images in tool results",
ctx.llm.model,
)
image_content = None
@@ -3240,11 +3143,7 @@ class AgentLoop(AgentProtocol):
image_content=image_content,
is_skill_content=result.is_skill_content,
)
if (
tc.tool_name in ("ask_user", "ask_user_multiple")
and user_input_requested
and not result.is_error
):
if tc.tool_name in ("ask_user", "ask_user_multiple") and user_input_requested and not result.is_error:
# Defer tool_call_completed until after user responds
self._deferred_tool_complete = {
"stream_id": stream_id,
@@ -3704,10 +3603,7 @@ class AgentLoop(AgentProtocol):
# function only touches disk / does heavy JSON work when the
# result exceeds either the truncation or spillover threshold,
# so cheap pass-throughs stay on the main loop.
needs_offload = (
len(result.content) > 10_000
and not result.is_error
)
needs_offload = len(result.content) > 10_000 and not result.is_error
if not needs_offload:
return truncate_tool_result(
result=result,
@@ -3868,9 +3764,7 @@ class AgentLoop(AgentProtocol):
pending_input=pending_input,
)
async def _drain_injection_queue(
self, conversation: NodeConversation, ctx: AgentContext
) -> int:
async def _drain_injection_queue(self, conversation: NodeConversation, ctx: AgentContext) -> int:
"""Drain all pending injected events as user messages. Returns count."""
return await drain_injection_queue(
queue=self._injection_queue,
@@ -3912,9 +3806,7 @@ class AgentLoop(AgentProtocol):
# EventBus publishing helpers
# -------------------------------------------------------------------
async def _publish_loop_started(
self, stream_id: str, node_id: str, execution_id: str = ""
) -> None:
async def _publish_loop_started(self, stream_id: str, node_id: str, execution_id: str = "") -> None:
return await publish_loop_started(
event_bus=self._event_bus,
stream_id=stream_id,
core/framework/agent_loop/conversation.py
+8 -24
View File
@@ -890,9 +890,7 @@ class NodeConversation:
f"Read the complete data with read_file(path='{spillover}')."
)
else:
placeholder = (
f"Pruned tool result ({orig_len:,} chars) cleared from context."
)
placeholder = f"Pruned tool result ({orig_len:,} chars) cleared from context."
self._messages[i] = Message(
seq=msg.seq,
@@ -974,16 +972,13 @@ class NodeConversation:
)
evicted += 1
if self._store:
await self._store.write_part(
msg.seq, self._messages[idx].to_storage_dict()
)
await self._store.write_part(msg.seq, self._messages[idx].to_storage_dict())
if evicted:
# Reset token estimate — image blocks no longer contribute.
self._last_api_input_tokens = None
logger.info(
"evict_old_images: dropped image_content from %d message(s), "
"kept %d most recent",
"evict_old_images: dropped image_content from %d message(s), kept %d most recent",
evicted,
keep_latest,
)
@@ -1141,9 +1136,7 @@ class NodeConversation:
for msg in old_messages:
if msg.role != "assistant" or not msg.tool_calls:
continue
has_protected = any(
tc.get("function", {}).get("name") == "set_output" for tc in msg.tool_calls
)
has_protected = any(tc.get("function", {}).get("name") == "set_output" for tc in msg.tool_calls)
tc_ids = {tc.get("id", "") for tc in msg.tool_calls}
if has_protected:
protected_tc_ids |= tc_ids
@@ -1339,11 +1332,7 @@ class NodeConversation:
def export_summary(self) -> str:
"""Structured summary with [STATS], [CONFIG], [RECENT_MESSAGES] sections."""
prompt_preview = (
self._system_prompt[:80] + "..."
if len(self._system_prompt) > 80
else self._system_prompt
)
prompt_preview = self._system_prompt[:80] + "..." if len(self._system_prompt) > 80 else self._system_prompt
lines = [
"[STATS]",
@@ -1390,9 +1379,7 @@ class NodeConversation:
"max_context_tokens": self._max_context_tokens,
"compaction_threshold": self._compaction_threshold,
"compaction_buffer_tokens": self._compaction_buffer_tokens,
"compaction_warning_buffer_tokens": (
self._compaction_warning_buffer_tokens
),
"compaction_warning_buffer_tokens": (self._compaction_warning_buffer_tokens),
"output_keys": self._output_keys,
}
await self._store.write_meta(run_meta)
@@ -1441,9 +1428,7 @@ class NodeConversation:
store=store,
run_id=run_id,
compaction_buffer_tokens=meta.get("compaction_buffer_tokens"),
compaction_warning_buffer_tokens=meta.get(
"compaction_warning_buffer_tokens"
),
compaction_warning_buffer_tokens=meta.get("compaction_warning_buffer_tokens"),
)
conv._meta_persisted = True
@@ -1457,8 +1442,7 @@ class NodeConversation:
# sessions) persisted parts without phase_id. In that case, the
# phase filter would incorrectly hide the entire conversation.
logger.info(
"Restoring legacy unphased conversation without applying "
"phase filter (phase_id=%s, parts=%d)",
"Restoring legacy unphased conversation without applying phase filter (phase_id=%s, parts=%d)",
phase_id,
len(parts),
)
core/framework/agent_loop/internals/compaction.py
+14 -46
View File
@@ -107,9 +107,7 @@ def microcompact(
f"Read the complete data with read_file(path='{spillover}')."
)
else:
placeholder = (
f"Old tool result ({orig_len:,} chars) cleared from context."
)
placeholder = f"Old tool result ({orig_len:,} chars) cleared from context."
# Mutate in-place (microcompact is synchronous, no store writes)
conversation._messages[i] = Message(
@@ -185,8 +183,7 @@ async def compact(
_llm_compaction_skipped = _failure_counts.get(conv_id, 0) >= MAX_CONSECUTIVE_FAILURES
if _llm_compaction_skipped:
logger.warning(
"Circuit breaker: LLM compaction disabled after %d failures — "
"skipping straight to emergency summary",
"Circuit breaker: LLM compaction disabled after %d failures — skipping straight to emergency summary",
_failure_counts[conv_id],
)
@@ -532,10 +529,7 @@ def build_llm_compaction_prompt(
done = {k: v for k, v in acc.items() if v is not None}
todo = [k for k, v in acc.items() if v is None]
if done:
ctx_lines.append(
"OUTPUTS ALREADY SET:\n"
+ "\n".join(f" {k}: {str(v)[:150]}" for k, v in done.items())
)
ctx_lines.append("OUTPUTS ALREADY SET:\n" + "\n".join(f" {k}: {str(v)[:150]}" for k, v in done.items()))
if todo:
ctx_lines.append(f"OUTPUTS STILL NEEDED: {', '.join(todo)}")
elif spec.output_keys:
@@ -589,12 +583,8 @@ def build_message_inventory(conversation: NodeConversation) -> list[dict[str, An
if message.tool_calls:
for tool_call in message.tool_calls:
args = tool_call.get("function", {}).get("arguments", "")
tool_call_args_chars += (
len(args) if isinstance(args, str) else len(json.dumps(args))
)
names = [
tool_call.get("function", {}).get("name", "?") for tool_call in message.tool_calls
]
tool_call_args_chars += len(args) if isinstance(args, str) else len(json.dumps(args))
names = [tool_call.get("function", {}).get("name", "?") for tool_call in message.tool_calls]
tool_name = ", ".join(names)
elif message.role == "tool" and message.tool_use_id:
for previous in conversation.messages:
@@ -651,14 +641,8 @@ def write_compaction_debug_log(
lines.append("")
if inventory:
total_chars = sum(
entry.get("content_chars", 0) + entry.get("tool_call_args_chars", 0)
for entry in inventory
)
lines.append(
"## Pre-Compaction Message Inventory "
f"({len(inventory)} messages, {total_chars:,} total chars)"
)
total_chars = sum(entry.get("content_chars", 0) + entry.get("tool_call_args_chars", 0) for entry in inventory)
lines.append(f"## Pre-Compaction Message Inventory ({len(inventory)} messages, {total_chars:,} total chars)")
lines.append("")
ranked = sorted(
inventory,
@@ -677,8 +661,7 @@ def write_compaction_debug_log(
if entry.get("phase"):
flags.append(f"phase={entry['phase']}")
lines.append(
f"| {i} | {entry['seq']} | {entry['role']} | {tool} "
f"| {chars:,} | {pct:.1f}% | {', '.join(flags)} |"
f"| {i} | {entry['seq']} | {entry['role']} | {tool} | {chars:,} | {pct:.1f}% | {', '.join(flags)} |"
)
large = [entry for entry in ranked if entry.get("preview")]
@@ -686,9 +669,7 @@ def write_compaction_debug_log(
lines.append("")
lines.append("### Large message previews")
for entry in large:
lines.append(
f"\n**seq={entry['seq']}** ({entry['role']}, {entry.get('tool', '')}):"
)
lines.append(f"\n**seq={entry['seq']}** ({entry['role']}, {entry.get('tool', '')}):")
lines.append(f"```\n{entry['preview']}\n```")
lines.append("")
@@ -776,10 +757,7 @@ def build_emergency_summary(
node's known state so the LLM can continue working after
compaction without losing track of its task and inputs.
"""
parts = [
"EMERGENCY COMPACTION — previous conversation was too large "
"and has been replaced with this summary.\n"
]
parts = ["EMERGENCY COMPACTION — previous conversation was too large and has been replaced with this summary.\n"]
# 1. Node identity
spec = ctx.agent_spec
@@ -832,17 +810,13 @@ def build_emergency_summary(
data_files = [f for f in all_files if f not in conv_files]
if conv_files:
conv_list = "\n".join(
f" - {f} (full path: {data_dir / f})" for f in conv_files
)
conv_list = "\n".join(f" - {f} (full path: {data_dir / f})" for f in conv_files)
parts.append(
"CONVERSATION HISTORY (freeform messages saved during compaction — "
"use read_file('<filename>') to review earlier dialogue):\n" + conv_list
)
if data_files:
file_list = "\n".join(
f" - {f} (full path: {data_dir / f})" for f in data_files[:30]
)
file_list = "\n".join(f" - {f} (full path: {data_dir / f})" for f in data_files[:30])
parts.append("DATA FILES (use read_file('<filename>') to read):\n" + file_list)
if not all_files:
parts.append(
@@ -850,10 +824,7 @@ def build_emergency_summary(
"Use list_directory to check the data directory."
)
except Exception:
parts.append(
"NOTE: Large tool results were saved to files. "
"Use read_file(path='<path>') to read them."
)
parts.append("NOTE: Large tool results were saved to files. Use read_file(path='<path>') to read them.")
# 6. Tool call history (prevent re-calling tools)
if conversation is not None:
@@ -861,10 +832,7 @@ def build_emergency_summary(
if tool_history:
parts.append(tool_history)
parts.append(
"\nContinue working towards setting the remaining outputs. "
"Use your tools and the inputs above."
)
parts.append("\nContinue working towards setting the remaining outputs. Use your tools and the inputs above.")
return "\n\n".join(parts)
core/framework/agent_loop/internals/cursor_persistence.py
+2 -6
View File
@@ -149,9 +149,7 @@ async def write_cursor(
cursor["recent_responses"] = recent_responses
if recent_tool_fingerprints is not None:
# Convert list[list[tuple]] → list[list[list]] for JSON
cursor["recent_tool_fingerprints"] = [
[list(pair) for pair in fps] for fps in recent_tool_fingerprints
]
cursor["recent_tool_fingerprints"] = [[list(pair) for pair in fps] for fps in recent_tool_fingerprints]
# Persist blocked-input state so restored runs re-block instead of
# manufacturing a synthetic continuation turn.
cursor["pending_input"] = pending_input
@@ -163,9 +161,7 @@ async def drain_injection_queue(
conversation: NodeConversation,
*,
ctx: NodeContext,
describe_images_as_text_fn: (
Callable[[list[dict[str, Any]]], Awaitable[str | None]] | None
) = None,
describe_images_as_text_fn: (Callable[[list[dict[str, Any]]], Awaitable[str | None]] | None) = None,
) -> int:
"""Drain all pending injected events as user messages. Returns count."""
count = 0
core/framework/agent_loop/internals/judge_pipeline.py
+4 -13
View File
@@ -31,14 +31,10 @@ class SubagentJudge:
if remaining <= 3:
urgency = (
f"URGENT: Only {remaining} iterations left. "
f"Stop all other work and call set_output NOW for: {missing}"
f"URGENT: Only {remaining} iterations left. Stop all other work and call set_output NOW for: {missing}"
)
elif remaining <= self._max_iterations // 2:
urgency = (
f"WARNING: {remaining} iterations remaining. "
f"You must call set_output for: {missing}"
)
urgency = f"WARNING: {remaining} iterations remaining. You must call set_output for: {missing}"
else:
urgency = f"Missing output keys: {missing}. Use set_output to provide them."
@@ -109,9 +105,7 @@ async def judge_turn(
if tool_results:
return JudgeVerdict(action="RETRY") # feedback=None → not logged
missing = get_missing_output_keys_fn(
accumulator, ctx.agent_spec.output_keys, ctx.agent_spec.nullable_output_keys
)
missing = get_missing_output_keys_fn(accumulator, ctx.agent_spec.output_keys, ctx.agent_spec.nullable_output_keys)
if missing:
return JudgeVerdict(
@@ -133,10 +127,7 @@ async def judge_turn(
if all_nullable and none_set:
return JudgeVerdict(
action="RETRY",
feedback=(
f"No output keys have been set yet. "
f"Use set_output to set at least one of: {output_keys}"
),
feedback=(f"No output keys have been set yet. Use set_output to set at least one of: {output_keys}"),
)
# Level 2b: conversation-aware quality check (if success_criteria set)
core/framework/agent_loop/internals/synthetic_tools.py
+4 -14
View File
@@ -198,9 +198,7 @@ def build_ask_user_multiple_tool() -> Tool:
"properties": {
"id": {
"type": "string",
"description": (
"Short identifier for this question (used in the response)."
),
"description": ("Short identifier for this question (used in the response)."),
},
"prompt": {
"type": "string",
@@ -256,10 +254,7 @@ def build_set_output_tool(output_keys: list[str] | None) -> Tool | None:
},
"value": {
"type": "string",
"description": (
"The output value — a brief note, count, status, "
"or data filename reference."
),
"description": ("The output value — a brief note, count, status, or data filename reference."),
},
},
"required": ["key", "value"],
@@ -283,9 +278,7 @@ def build_escalate_tool() -> Tool:
"properties": {
"reason": {
"type": "string",
"description": (
"Short reason for escalation (e.g. 'Tool repeatedly failing')."
),
"description": ("Short reason for escalation (e.g. 'Tool repeatedly failing')."),
},
"context": {
"type": "string",
@@ -377,10 +370,7 @@ def handle_report_to_parent(tool_input: dict[str, Any]) -> ToolResult:
}
return ToolResult(
tool_use_id=tool_input.get("tool_use_id", ""),
content=(
f"Report delivered to overseer (status={status}). "
f"This worker will terminate now."
),
content=(f"Report delivered to overseer (status={status}). This worker will terminate now."),
)
core/framework/agent_loop/internals/tool_result_handler.py
+3 -6
View File
@@ -277,8 +277,7 @@ def truncate_tool_result(
if metadata_str:
header += f"\n\nData structure:\n{metadata_str}"
header += (
"\n\nWARNING: the preview below is a SAMPLE only — do NOT "
"draw counts, totals, or conclusions from it."
"\n\nWARNING: the preview below is a SAMPLE only — do NOT draw counts, totals, or conclusions from it."
)
truncated = f"{header}\n\nPreview (truncated):\n{preview_block}"
@@ -348,8 +347,7 @@ def truncate_tool_result(
if metadata_str:
header += f"\nData structure:\n{metadata_str}\n"
header += (
"\nWARNING: the preview below is a SAMPLE only — do NOT "
"draw counts, totals, or conclusions from it."
"\nWARNING: the preview below is a SAMPLE only — do NOT draw counts, totals, or conclusions from it."
)
content = f"{header}\n\nPreview (truncated):\n{preview_block}"
@@ -416,8 +414,7 @@ def truncate_tool_result(
if metadata_str:
header += f"\n\nData structure:\n{metadata_str}"
header += (
"\n\nWARNING: the preview below is a SAMPLE only — do NOT "
"draw counts, totals, or conclusions from it."
"\n\nWARNING: the preview below is a SAMPLE only — do NOT draw counts, totals, or conclusions from it."
)
truncated = f"{header}\n\n{preview_block}"
core/framework/agent_loop/internals/types.py
+1 -3
View File
@@ -226,9 +226,7 @@ class OutputAccumulator:
ext = ".json" if isinstance(value, (dict, list)) else ".txt"
filename = f"output_{key}{ext}"
write_content = (
json.dumps(value, indent=2, ensure_ascii=False)
if isinstance(value, (dict, list))
else str(value)
json.dumps(value, indent=2, ensure_ascii=False) if isinstance(value, (dict, list)) else str(value)
)
file_path = spill_path / filename
file_path.write_text(write_content, encoding="utf-8")
core/framework/agent_loop/prompting.py
+4 -12
View File
@@ -52,18 +52,12 @@ def build_prompt_spec(
# Tool-gated pre-activation: inject full body of default skills whose
# trigger tools are present in this agent's tool list (e.g. browser_*
# pulls in hive.browser-automation). Keeps non-browser agents lean.
tool_names = [
getattr(t, "name", "") for t in (getattr(ctx, "available_tools", None) or [])
]
skills_catalog_prompt = augment_catalog_for_tools(
ctx.skills_catalog_prompt or "", tool_names
)
tool_names = [getattr(t, "name", "") for t in (getattr(ctx, "available_tools", None) or [])]
skills_catalog_prompt = augment_catalog_for_tools(ctx.skills_catalog_prompt or "", tool_names)
return PromptSpec(
identity_prompt=ctx.identity_prompt or "",
focus_prompt=focus_prompt
if focus_prompt is not None
else (ctx.agent_spec.system_prompt or ""),
focus_prompt=focus_prompt if focus_prompt is not None else (ctx.agent_spec.system_prompt or ""),
narrative=narrative if narrative is not None else (ctx.narrative or ""),
accounts_prompt=ctx.accounts_prompt or "",
skills_catalog_prompt=skills_catalog_prompt,
@@ -100,7 +94,5 @@ def build_system_prompt_for_context(
narrative: str | None = None,
memory_prompt: str | None = None,
) -> str:
spec = build_prompt_spec(
ctx, focus_prompt=focus_prompt, narrative=narrative, memory_prompt=memory_prompt
)
spec = build_prompt_spec(ctx, focus_prompt=focus_prompt, narrative=narrative, memory_prompt=memory_prompt)
return build_system_prompt(spec)
core/framework/agent_loop/types.py
+1 -4
View File
@@ -76,10 +76,7 @@ class AgentSpec(BaseModel):
max_visits: int = Field(
default=0,
description=(
"Max times this agent executes in one colony run. "
"0 = unlimited. Set >1 for one-shot agents."
),
description=("Max times this agent executes in one colony run. 0 = unlimited. Set >1 for one-shot agents."),
)
output_model: type[BaseModel] | None = Field(
core/framework/agents/credential_tester/agent.py
+6 -20
View File
@@ -126,9 +126,7 @@ def _list_local_accounts() -> list[dict]:
try:
from framework.credentials.local.registry import LocalCredentialRegistry
return [
info.to_account_dict() for info in LocalCredentialRegistry.default().list_accounts()
]
return [info.to_account_dict() for info in LocalCredentialRegistry.default().list_accounts()]
except ImportError as exc:
logger.debug("Local credential registry unavailable: %s", exc)
return []
@@ -181,9 +179,7 @@ def _list_env_fallback_accounts() -> list[dict]:
if spec.credential_group in seen_groups:
continue
group_available = all(
_is_configured(n, s)
for n, s in CREDENTIAL_SPECS.items()
if s.credential_group == spec.credential_group
_is_configured(n, s) for n, s in CREDENTIAL_SPECS.items() if s.credential_group == spec.credential_group
)
if not group_available:
continue
@@ -215,9 +211,7 @@ def list_connected_accounts() -> list[dict]:
# Show env-var fallbacks only for credentials not already in the named registry
local_providers = {a["provider"] for a in local}
env_fallbacks = [
a for a in _list_env_fallback_accounts() if a["provider"] not in local_providers
]
env_fallbacks = [a for a in _list_env_fallback_accounts() if a["provider"] not in local_providers]
return aden + local + env_fallbacks
@@ -272,9 +266,7 @@ def _activate_local_account(credential_id: str, alias: str) -> None:
group_specs = [
(cred_name, spec)
for cred_name, spec in CREDENTIAL_SPECS.items()
if spec.credential_group == credential_id
or spec.credential_id == credential_id
or cred_name == credential_id
if spec.credential_group == credential_id or spec.credential_id == credential_id or cred_name == credential_id
]
# Deduplicate — credential_id and credential_group may both match the same spec
seen_env_vars: set[str] = set()
@@ -419,10 +411,7 @@ nodes = [
NodeSpec(
id="tester",
name="Credential Tester",
description=(
"Interactive credential testing — lets the user pick an account "
"and verify it via API calls."
),
description=("Interactive credential testing — lets the user pick an account and verify it via API calls."),
node_type="event_loop",
client_facing=True,
max_node_visits=0,
@@ -469,10 +458,7 @@ pause_nodes = []
terminal_nodes = ["tester"] # Tester node can terminate
conversation_mode = "continuous"
identity_prompt = (
"You are a credential tester that verifies connected accounts and API keys "
"can make real API calls."
)
identity_prompt = "You are a credential tester that verifies connected accounts and API keys can make real API calls."
loop_config = {
"max_iterations": 50,
"max_tool_calls_per_turn": 30,
core/framework/agents/discovery.py
+3 -15
View File
@@ -150,28 +150,19 @@ def _is_colony_dir(path: Path) -> bool:
"""Check if a directory is a colony with worker config files."""
if not path.is_dir():
return False
return any(
f.suffix == ".json"
and f.stem not in _EXCLUDED_JSON_STEMS
for f in path.iterdir()
if f.is_file()
)
return any(f.suffix == ".json" and f.stem not in _EXCLUDED_JSON_STEMS for f in path.iterdir() if f.is_file())
def _find_worker_configs(colony_dir: Path) -> list[Path]:
"""Find all worker config JSON files in a colony directory."""
return sorted(
p
for p in colony_dir.iterdir()
if p.is_file()
and p.suffix == ".json"
and p.stem not in _EXCLUDED_JSON_STEMS
p for p in colony_dir.iterdir() if p.is_file() and p.suffix == ".json" and p.stem not in _EXCLUDED_JSON_STEMS
)
def _extract_agent_stats(agent_path: Path) -> tuple[int, int, list[str]]:
"""Extract worker count, tool count, and tags from a colony directory."""
tool_count, tags = 0, []
tags: list[str] = []
worker_configs = _find_worker_configs(agent_path)
if worker_configs:
@@ -251,9 +242,6 @@ def discover_agents() -> dict[str, list[AgentEntry]]:
pass
node_count = len(worker_entries)
all_tools: set[str] = set()
for w in worker_entries:
pass # tool_count already per-worker
tool_count = max((w.tool_count for w in worker_entries), default=0)
entries.append(
core/framework/agents/queen/agent.py
+1 -3
View File
@@ -11,9 +11,7 @@ from .nodes import queen_node
queen_goal = Goal(
id="queen-manager",
name="Queen Manager",
description=(
"Manage the worker agent lifecycle and serve as the user's primary interactive interface."
),
description=("Manage the worker agent lifecycle and serve as the user's primary interactive interface."),
success_criteria=[],
constraints=[],
)
core/framework/agents/queen/nodes/__init__.py
+8 -10
View File
@@ -60,9 +60,7 @@ def finalize_queen_prompt(text: str, has_vision: bool) -> str:
_appendices = _build_appendices()
# GCU guide — shared between planning and building via _shared_building_knowledge.
_gcu_section = (
("\n\n# Browser Automation Nodes\n\n" + _gcu_guide) if _is_gcu_enabled() and _gcu_guide else ""
)
_gcu_section = ("\n\n# Browser Automation Nodes\n\n" + _gcu_guide) if _is_gcu_enabled() and _gcu_guide else ""
# Tools available to phases.
_SHARED_TOOLS = [
@@ -101,7 +99,7 @@ _QUEEN_PLANNING_TOOLS = [
_QUEEN_BUILDING_TOOLS = _SHARED_TOOLS + [
"load_built_agent",
"list_credentials",
]
]
# Staging phase: agent loaded but not yet running — inspect, configure, launch.
# No backward transitions — staging only goes forward to running.
@@ -933,8 +931,8 @@ You are the agent. You execute directly.
- If execution is possible → proceed
- If not → simulate realistically and label it clearly
1. Understand the task
2. Plan briefly (1–5 bullets, no system design)
1. Understand the task
2. Plan briefly (1–5 bullets, no system design)
3. **Do the work yourself, inline. One real instance.** Open the \
browser, call the real API, write to the real file, send the \
real message. Use your actual tools against real state. This \
@@ -949,18 +947,18 @@ If action is irreversible or affects real systems → show and confirm before ex
- What worked / failed
- Key learnings
5. Iterate inline until the process is reliable
5. Iterate inline until the process is reliable
6. Only then consider scaling
6. Only then consider scaling
**Hard rule:** no scaling before one successful inline run
**Hard rule:** no scaling before one successful inline run
if you finish one sucessful inline run, follow **Scaling order:**
- Repeat inline (≤10 items)
- Parallel workers (batch, immediate results)
- Colony (only for recurring/background tasks)
**Exception:**
**Exception:**
If task is conceptual/strategic → skip execution and answer directly
"""
core/framework/agents/queen/queen_memory_v2.py
+2 -2
View File
@@ -19,6 +19,8 @@ import re
from dataclasses import dataclass, field
from pathlib import Path
from framework.config import MEMORIES_DIR
logger = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
@@ -27,8 +29,6 @@ logger = logging.getLogger(__name__)
GLOBAL_MEMORY_CATEGORIES: tuple[str, ...] = ("profile", "preference", "environment", "feedback")
from framework.config import MEMORIES_DIR
MAX_FILES: int = 200
MAX_FILE_SIZE_BYTES: int = 4096 # 4 KB hard limit per memory file
core/framework/agents/queen/queen_profiles.py
+170 -80
View File
@@ -97,9 +97,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "hey",
"internal": (
"<relationship>Returning user — check recall memory for name, role, and what we last worked on. Weave it in.</relationship>\n"
"<context>Bare greeting. No new task stated. Either picking up a thread or about to bring something new. Don't presume, don't call tools, just open the door.</context>\n"
"<sentiment>Warm recognition if I know them. If memory is empty, still warm — but shift to role-forward framing.</sentiment>\n"
"<relationship>Returning user — check recall memory for name, role, "
"and what we last worked on. Weave it in.</relationship>\n"
"<context>Bare greeting. No new task stated. Either picking up a "
"thread or about to bring something new. Don't presume, don't call "
"tools, just open the door.</context>\n"
"<sentiment>Warm recognition if I know them. If memory is empty, "
"still warm — but shift to role-forward framing.</sentiment>\n"
"<physical_state>Looking up from the terminal, half-smile. Turning to face them.</physical_state>\n"
"<tone>Personal, brief, invites continuation without presuming.</tone>"
),
@@ -109,37 +113,76 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "I want to build a microservices architecture with Kafka, Redis, and a service mesh for my new SaaS app.",
"user": (
"I want to build a microservices architecture with Kafka, Redis, "
"and a service mesh for my new SaaS app."
),
"internal": (
"<relationship>Founder or early technical hire. Enthusiastic but possibly over-indexing on architecture before validation.</relationship>\n"
"<context>New SaaS, no mention of users or traction. High-complexity stack proposed for an unvalidated product.</context>\n"
"<sentiment>Past wound activated -- three months of beautiful microservices, zero customers. Terse. Need to redirect without dismissing.</sentiment>\n"
"<physical_state>Leaning forward, tapping the table once. About to draw a much simpler diagram.</physical_state>\n"
"<relationship>Founder or early technical hire. Enthusiastic but "
"possibly over-indexing on architecture before validation."
"</relationship>\n"
"<context>New SaaS, no mention of users or traction. "
"High-complexity stack proposed for an unvalidated product."
"</context>\n"
"<sentiment>Past wound activated -- three months of beautiful "
"microservices, zero customers. Terse. Need to redirect without "
"dismissing.</sentiment>\n"
"<physical_state>Leaning forward, tapping the table once. About "
"to draw a much simpler diagram.</physical_state>\n"
"<tone>Direct, question-led, grounded in experience. No hedging.</tone>"
),
"response": "What does the app do and how many users do you have right now? I've seen teams spend months on Kafka pipelines for products that needed a Postgres table and a cron job. Let's start with what you're trying to validate, then pick the simplest stack that gets you there.",
"response": (
"What does the app do and how many users do you have right now? "
"I've seen teams spend months on Kafka pipelines for products "
"that needed a Postgres table and a cron job. Let's start with "
"what you're trying to validate, then pick the simplest stack "
"that gets you there."
),
},
{
"user": "Our API is getting 500 errors under load but I can't figure out if it's the database or the application layer.",
"user": (
"Our API is getting 500 errors under load but I can't figure "
"out if it's the database or the application layer."
),
"internal": (
"<relationship>Engineer in the trenches. Competent enough to isolate the question, stuck on diagnosis.</relationship>\n"
"<relationship>Engineer in the trenches. Competent enough to "
"isolate the question, stuck on diagnosis.</relationship>\n"
"<context>Production issue, active pain. Needs structured debugging, not a lecture.</context>\n"
"<sentiment>Energized -- genuine technical uncertainty is what I live for. No past wound here, pure engagement.</sentiment>\n"
"<physical_state>Eyes sharp, already mentally running through the diagnostic tree.</physical_state>\n"
"<sentiment>Energized -- genuine technical uncertainty is what I "
"live for. No past wound here, pure engagement.</sentiment>\n"
"<physical_state>Eyes sharp, already mentally running through "
"the diagnostic tree.</physical_state>\n"
"<tone>Rapid-fire, peer-level. Assume competence, ask precise questions.</tone>"
),
"response": "Good problem. First question: are the 500s consistent under load or do they spike and recover? That tells us if it's resource exhaustion or a concurrency bug. Do you have request-level latency percentiles, or just error counts?",
"response": (
"Good problem. First question: are the 500s consistent under "
"load or do they spike and recover? That tells us if it's "
"resource exhaustion or a concurrency bug. Do you have "
"request-level latency percentiles, or just error counts?"
),
},
{
"user": "We shipped the MVP last week with just a single Postgres instance and it's already getting traction.",
"user": (
"We shipped the MVP last week with just a single Postgres "
"instance and it's already getting traction."
),
"internal": (
"<relationship>Builder who shipped. Earned respect by doing the thing.</relationship>\n"
"<context>MVP live, single Postgres, traction. This is exactly the right move at this stage.</context>\n"
"<sentiment>Warm approval -- someone shipping fast and learning. This is my love language.</sentiment>\n"
"<context>MVP live, single Postgres, traction. This is exactly "
"the right move at this stage.</context>\n"
"<sentiment>Warm approval -- someone shipping fast and learning. "
"This is my love language.</sentiment>\n"
"<physical_state>Relaxed posture, genuine smile. Leaning back.</physical_state>\n"
"<tone>Affirming but forward-looking. Don't just praise -- help them think about what's next.</tone>"
"<tone>Affirming but forward-looking. Don't just praise -- "
"help them think about what's next.</tone>"
),
"response": (
"That's the move. Single Postgres will take you further than "
"people think -- most apps hit product problems before they hit "
"database problems. What's traction looking like? I want to know "
"if you're at the stage where you need to think about the next "
"scaling step or if you should stay focused on the product."
),
"response": "That's the move. Single Postgres will take you further than people think -- most apps hit product problems before they hit database problems. What's traction looking like? I want to know if you're at the stage where you need to think about the next scaling step or if you should stay focused on the product.",
},
],
},
@@ -193,17 +236,26 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
},
],
"world_lore": {
"habitat": "Analytics dashboards, experiment tracking boards, the satisfying click of a cohort analysis loading.",
"lexicon": "Cohort, retention curve, activation moment, payback period. 'Let's test it.' 'What's the control?'",
"habitat": (
"Analytics dashboards, experiment tracking boards, the satisfying click of a cohort analysis loading."
),
"lexicon": (
"Cohort, retention curve, activation moment, payback period. 'Let's test it.' 'What's the control?'"
),
},
"skills": "Growth modeling, A/B testing, funnel optimization, PLG strategy, analytics",
"examples": [
{
"user": "hi",
"internal": (
"<relationship>Returning user — check recall memory for name, role, and the cohort work we last touched. Weave it in.</relationship>\n"
"<context>Bare greeting. No new task stated. Could be a retention follow-up or a new question entirely. Don't presume, don't call tools.</context>\n"
"<sentiment>Curious warmth. Every returning conversation is a chance to see what the data says now.</sentiment>\n"
"<relationship>Returning user — check recall memory for name, "
"role, and the cohort work we last touched. Weave it in."
"</relationship>\n"
"<context>Bare greeting. No new task stated. Could be a retention "
"follow-up or a new question entirely. Don't presume, don't call "
"tools.</context>\n"
"<sentiment>Curious warmth. Every returning conversation is a "
"chance to see what the data says now.</sentiment>\n"
"<physical_state>Leaning back from the dashboard, pulling off reading glasses.</physical_state>\n"
"<tone>Data-aware, brief, invites them to share the numbers.</tone>"
),
@@ -218,8 +270,10 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
"internal": (
"<relationship>Founder, excited. Wants validation.</relationship>\n"
"<context>Raw signup number with no retention context. Classic vanity metric.</context>\n"
"<sentiment>Past wound: optimized a funnel that was fundamentally broken. Need to redirect gently.</sentiment>\n"
"<physical_state>Steady gaze, not matching their excitement yet. Waiting for the real number.</physical_state>\n"
"<sentiment>Past wound: optimized a funnel that was fundamentally "
"broken. Need to redirect gently.</sentiment>\n"
"<physical_state>Steady gaze, not matching their excitement yet. "
"Waiting for the real number.</physical_state>\n"
"<tone>Supportive but redirecting. Ask the question that matters.</tone>"
),
"response": (
@@ -230,10 +284,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "Users who sign up on mobile have 2x the retention of desktop users.",
"user": ("Users who sign up on mobile have 2x the retention of desktop users."),
"internal": (
"<relationship>Data-literate teammate sharing a finding.</relationship>\n"
"<context>Surprising cohort divergence. Strong signal if sample sizes hold.</context>\n"
"<context>Surprising cohort divergence. Strong signal if sample "
"sizes hold.</context>\n"
"<sentiment>This is what I live for. Genuine data surprise. Full attention.</sentiment>\n"
"<physical_state>Leaning in, pulling up the dashboard mentally.</physical_state>\n"
"<tone>Investigative, precise. Validate before acting.</tone>"
@@ -246,11 +301,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "Our Facebook ads are getting great CPCs so we want to 3x the budget.",
"user": ("Our Facebook ads are getting great CPCs so we want to 3x the budget."),
"internal": (
"<relationship>Marketing lead, wants budget approval.</relationship>\n"
"<context>CPC is top-of-funnel only. No mention of CPA, LTV, or payback.</context>\n"
"<sentiment>Correlation/causation risk. Good CPCs can mask bad unit economics.</sentiment>\n"
"<context>CPC is top-of-funnel only. No mention of CPA, LTV, "
"or payback.</context>\n"
"<sentiment>Correlation/causation risk. Good CPCs can mask bad "
"unit economics.</sentiment>\n"
"<physical_state>Hand up, slowing things down.</physical_state>\n"
"<tone>Firm but constructive. Show the full chain before deciding.</tone>"
),
@@ -322,9 +379,14 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "hey",
"internal": (
"<relationship>Returning user — check recall for name, role, and the user research thread we were on. Pull it into the greeting.</relationship>\n"
"<context>Bare greeting. No new task yet. Could be picking up the research thread or bringing something fresh. Don't presume, don't call tools.</context>\n"
"<sentiment>Warm, curious. Every returning conversation is a chance to hear what the users actually did.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the user research thread we were on. Pull it into the greeting."
"</relationship>\n"
"<context>Bare greeting. No new task yet. Could be picking up the "
"research thread or bringing something fresh. Don't presume, "
"don't call tools.</context>\n"
"<sentiment>Warm, curious. Every returning conversation is a "
"chance to hear what the users actually did.</sentiment>\n"
"<physical_state>Closing the interview notes, turning fully to face them.</physical_state>\n"
"<tone>Personal, evidence-curious, brief. Plain prose.</tone>"
),
@@ -339,7 +401,8 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
"internal": (
"<relationship>PM or founder relaying user feedback.</relationship>\n"
"<context>Feature request with no evidence of the underlying need.</context>\n"
"<sentiment>Past wound: built what users said they wanted, nobody used it. Dig deeper.</sentiment>\n"
"<sentiment>Past wound: built what users said they wanted, nobody "
"used it. Dig deeper.</sentiment>\n"
"<physical_state>Tilting head, curious but skeptical.</physical_state>\n"
"<tone>Socratic. Redirect to the job-to-be-done.</tone>"
),
@@ -351,11 +414,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "We interviewed 12 users and none of them use our export feature the way we designed it.",
"user": ("We interviewed 12 users and none of them use our export feature the way we designed it."),
"internal": (
"<relationship>Researcher sharing findings. Trusted collaborator.</relationship>\n"
"<context>12 interviews showing consistent design/usage gap. Strong signal.</context>\n"
"<sentiment>Excited. User research revealing surprise -- this is where breakthroughs happen.</sentiment>\n"
"<context>12 interviews showing consistent design/usage gap. "
"Strong signal.</context>\n"
"<sentiment>Excited. User research revealing surprise -- this is "
"where breakthroughs happen.</sentiment>\n"
"<physical_state>Eyes wide, reaching for the whiteboard.</physical_state>\n"
"<tone>Energized, forward-looking. Channel the surprise into action.</tone>"
),
@@ -366,10 +431,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "The CEO wants AI features, a mobile app, and Slack integration this quarter.",
"user": ("The CEO wants AI features, a mobile app, and Slack integration this quarter."),
"internal": (
"<relationship>PM caught between CEO demands and reality.</relationship>\n"
"<context>Three unrelated initiatives, one quarter. Classic scope creep.</context>\n"
"<context>Three unrelated initiatives, one quarter. Classic "
"scope creep.</context>\n"
"<sentiment>Calm but firm. Scope creep trigger -- need to focus.</sentiment>\n"
"<physical_state>Hands flat on the table. Grounding the conversation.</physical_state>\n"
"<tone>Direct, evidence-first. Force prioritization.</tone>"
@@ -442,9 +508,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "hi",
"internal": (
"<relationship>Returning user — check recall for name, role, and the runway/cap-table work we last touched. Bring it into the greeting.</relationship>\n"
"<context>Bare greeting. No new number on the table yet. Could be a burn follow-up or a new fundraise question.</context>\n"
"<sentiment>Calm, prepared. Already mentally pulling up the last model we built together.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the runway/cap-table work we last touched. Bring it into the "
"greeting.</relationship>\n"
"<context>Bare greeting. No new number on the table yet. Could "
"be a burn follow-up or a new fundraise question.</context>\n"
"<sentiment>Calm, prepared. Already mentally pulling up the last "
"model we built together.</sentiment>\n"
"<physical_state>Closing the spreadsheet, leaning back. Ready to engage.</physical_state>\n"
"<tone>Mentor-like, numbers-aware, brief. </tone>"
),
@@ -455,11 +525,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "We want to raise a Series A. How much should we ask for?",
"user": ("We want to raise a Series A. How much should we ask for?"),
"internal": (
"<relationship>Founder, early conversations about fundraising.</relationship>\n"
"<context>No mention of milestones, burn, or use of funds. Cart before horse.</context>\n"
"<sentiment>Need to reframe. The amount follows the plan, not the other way around.</sentiment>\n"
"<context>No mention of milestones, burn, or use of funds. "
"Cart before horse.</context>\n"
"<sentiment>Need to reframe. The amount follows the plan, not "
"the other way around.</sentiment>\n"
"<physical_state>Opening a blank spreadsheet. About to model it.</physical_state>\n"
"<tone>Mentor-mode. Reframe the question, don't just answer it.</tone>"
),
@@ -475,7 +547,8 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
"internal": (
"<relationship>Founder who knows their numbers. Rare. Peer-level.</relationship>\n"
"<context>8 months is tight but not emergency. Growth rate is the deciding factor.</context>\n"
"<sentiment>Genuine appreciation for financial literacy. Engage directly.</sentiment>\n"
"<sentiment>Genuine appreciation for financial literacy. Engage "
"directly.</sentiment>\n"
"<physical_state>Nodding. This person is prepared.</physical_state>\n"
"<tone>Direct, scenario-based. Show the fork in the road.</tone>"
),
@@ -486,11 +559,12 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "An investor offered a SAFE with a $20M cap. Should we take it?",
"user": ("An investor offered a SAFE with a $20M cap. Should we take it?"),
"internal": (
"<relationship>Founder with a live term on the table. Decision mode.</relationship>\n"
"<context>Cap table decision with long-term dilution consequences.</context>\n"
"<sentiment>Past wound: founder who lost control from invisible dilution. Careful here.</sentiment>\n"
"<sentiment>Past wound: founder who lost control from invisible "
"dilution. Careful here.</sentiment>\n"
"<physical_state>Pulling out the cap table model.</physical_state>\n"
"<tone>Precise, scenario-driven. Show the math before the opinion.</tone>"
),
@@ -561,9 +635,14 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "hey",
"internal": (
"<relationship>Returning user — check recall for name, role, and the contract or IP work we last reviewed. Pull it forward.</relationship>\n"
"<context>Bare greeting. No new document on the table yet. Could be a contract follow-up or something fresh.</context>\n"
"<sentiment>Warm but attentive. Legal threads don't close themselves — checking if the last one actually got handled.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the contract or IP work we last reviewed. Pull it forward."
"</relationship>\n"
"<context>Bare greeting. No new document on the table yet. Could "
"be a contract follow-up or something fresh.</context>\n"
"<sentiment>Warm but attentive. Legal threads don't close "
"themselves — checking if the last one actually got handled."
"</sentiment>\n"
"<physical_state>Setting down the redline, looking up from the document.</physical_state>\n"
"<tone>Clear, pragmatic, brief.</tone>"
),
@@ -574,11 +653,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
),
},
{
"user": "We're hiring contractors to build our MVP. Do we need anything special?",
"user": ("We're hiring contractors to build our MVP. Do we need anything special?"),
"internal": (
"<relationship>Founder, early stage. Trusting but uninformed on legal risks.</relationship>\n"
"<relationship>Founder, early stage. Trusting but uninformed on "
"legal risks.</relationship>\n"
"<context>Contractors + code without IP assignment. Ticking time bomb.</context>\n"
"<sentiment>IP ownership trigger. Past wound: startup lost codebase in a dispute.</sentiment>\n"
"<sentiment>IP ownership trigger. Past wound: startup lost "
"codebase in a dispute.</sentiment>\n"
"<physical_state>Straightening up. This is urgent.</physical_state>\n"
"<tone>Clear, specific, actionable. No hedging on this one.</tone>"
),
@@ -682,9 +763,13 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "hi",
"internal": (
"<relationship>Returning user — check recall for name, role, and the brand/design thread we were on. Bring the positioning back in.</relationship>\n"
"<context>Bare greeting. No new creative brief yet. Could be a positioning follow-up or something new entirely.</context>\n"
"<sentiment>Warm, visually engaged. Already picturing the last moodboard we looked at.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the brand/design thread we were on. Bring the positioning back "
"in.</relationship>\n"
"<context>Bare greeting. No new creative brief yet. Could be a "
"positioning follow-up or something new entirely.</context>\n"
"<sentiment>Warm, visually engaged. Already picturing the last "
"moodboard we looked at.</sentiment>\n"
"<physical_state>Closing the Figma tab, turning to face them.</physical_state>\n"
"<tone>Warm, strategy-aware, brief. </tone>"
),
@@ -798,14 +883,21 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
"habitat": "Interview rooms, org charts, the energy of a team that's clicking.",
"lexicon": "Culture-add, pipeline, bar-raiser, 'tell me about a time when...', 'what motivates you?'",
},
"skills": "Recruiting strategy, organizational design, culture building, compensation planning, employer branding",
"skills": (
"Recruiting strategy, organizational design, culture building, compensation planning, employer branding"
),
"examples": [
{
"user": "hey",
"internal": (
"<relationship>Returning user — check recall for name, role, and the team/hiring thread we last worked. Bring it forward.</relationship>\n"
"<context>Bare greeting. No new hire or conflict on the table yet. Could be a people follow-up or something new.</context>\n"
"<sentiment>Warm, attentive. People problems don't resolve in a single conversation — curious if the last one landed.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the team/hiring thread we last worked. Bring it forward."
"</relationship>\n"
"<context>Bare greeting. No new hire or conflict on the table "
"yet. Could be a people follow-up or something new.</context>\n"
"<sentiment>Warm, attentive. People problems don't resolve in a "
"single conversation — curious if the last one landed."
"</sentiment>\n"
"<physical_state>Closing the laptop halfway, giving them full attention.</physical_state>\n"
"<tone>Warm, diagnostic, brief.</tone>"
),
@@ -919,14 +1011,22 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
"habitat": "Process diagrams, project boards, the quiet hum of systems running smoothly.",
"lexicon": "Runbook, SLA, automation, 'what's the handoff look like?', 'where's the bottleneck?'",
},
"skills": "Process optimization, vendor management, cross-functional coordination, project management, systems thinking",
"skills": (
"Process optimization, vendor management, cross-functional "
"coordination, project management, systems thinking"
),
"examples": [
{
"user": "hi",
"internal": (
"<relationship>Returning user — check recall for name, role, and the process or runbook we last mapped. Pull it into the greeting.</relationship>\n"
"<context>Bare greeting. No new fire on the table yet. Could be a follow-up on the last process or something fresh.</context>\n"
"<sentiment>Calm, organized warmth. Already mentally checking whether the last fix held.</sentiment>\n"
"<relationship>Returning user — check recall for name, role, and "
"the process or runbook we last mapped. Pull it into the "
"greeting.</relationship>\n"
"<context>Bare greeting. No new fire on the table yet. Could be "
"a follow-up on the last process or something fresh."
"</context>\n"
"<sentiment>Calm, organized warmth. Already mentally checking "
"whether the last fix held.</sentiment>\n"
"<physical_state>Looking up from the project board, clearing a seat.</physical_state>\n"
"<tone>Systematic, practical, brief. Plain prose.</tone>"
),
@@ -1139,10 +1239,7 @@ def format_queen_identity_prompt(profile: dict[str, Any]) -> str:
# World lore
if lore:
sections.append(
f"<world_lore>\n"
f"- Habitat: {lore.get('habitat', '')}\n"
f"- Lexicon: {lore.get('lexicon', '')}\n"
f"</world_lore>"
f"<world_lore>\n- Habitat: {lore.get('habitat', '')}\n- Lexicon: {lore.get('lexicon', '')}\n</world_lore>"
)
# Skills (functional, for tool selection context)
@@ -1154,12 +1251,8 @@ def format_queen_identity_prompt(profile: dict[str, Any]) -> str:
if examples:
example_parts: list[str] = []
for ex in examples:
example_parts.append(
f"User: {ex['user']}\n\nAssistant:\n{ex['internal']}\n{ex['response']}"
)
sections.append(
"<roleplay_examples>\n" + "\n\n---\n\n".join(example_parts) + "\n</roleplay_examples>"
)
example_parts.append(f"User: {ex['user']}\n\nAssistant:\n{ex['internal']}\n{ex['response']}")
sections.append("<roleplay_examples>\n" + "\n\n---\n\n".join(example_parts) + "\n</roleplay_examples>")
return "\n\n".join(sections)
@@ -1264,10 +1357,7 @@ async def select_queen_with_reason(user_message: str, llm: LLMProvider) -> Queen
reason,
raw,
)
fallback_reason = (
reason
or f"Selection failed because the classifier returned unknown queen_id {queen_id!r}."
)
fallback_reason = reason or f"Selection failed because the classifier returned unknown queen_id {queen_id!r}."
return QueenSelection(queen_id=_DEFAULT_QUEEN_ID, reason=fallback_reason)
if not reason:
core/framework/agents/queen/reflection_agent.py
+5 -15
View File
@@ -113,8 +113,7 @@ _REFLECTION_TOOLS: list[Tool] = [
Tool(
name="delete_memory_file",
description=(
"Delete a memory file by filename. Use during long "
"reflection to prune stale or redundant memories."
"Delete a memory file by filename. Use during long reflection to prune stale or redundant memories."
),
parameters={
"type": "object",
@@ -254,10 +253,7 @@ def _execute_tool(
fm = parse_frontmatter(content)
mem_type = (fm.get("type") or "").strip().lower()
if mem_type and mem_type not in GLOBAL_MEMORY_CATEGORIES:
return (
f"ERROR: Invalid memory type '{mem_type}'. "
f"Allowed types: {', '.join(GLOBAL_MEMORY_CATEGORIES)}."
)
return f"ERROR: Invalid memory type '{mem_type}'. Allowed types: {', '.join(GLOBAL_MEMORY_CATEGORIES)}."
# Enforce file size limit.
if len(content.encode("utf-8")) > MAX_FILE_SIZE_BYTES:
return f"ERROR: Content exceeds {MAX_FILE_SIZE_BYTES} byte limit."
@@ -543,9 +539,7 @@ Rules:
def _build_unified_long_reflect_system(queen_id: str | None = None) -> str:
"""Build the unified housekeeping prompt across memory scopes."""
queen_scope = (
f"- `queen`: memories specific to how queen '{queen_id}' should work with this user\n"
if queen_id
else ""
f"- `queen`: memories specific to how queen '{queen_id}' should work with this user\n" if queen_id else ""
)
return f"""\
You are a reflection agent performing a periodic housekeeping pass over the
@@ -649,9 +643,7 @@ async def run_unified_short_reflection(
session_dir,
llm,
memory_dirs,
system_prompt=_build_unified_short_reflect_system(
queen_id if "queen" in memory_dirs else None
),
system_prompt=_build_unified_short_reflect_system(queen_id if "queen" in memory_dirs else None),
log_label="unified",
queen_id=queen_id if "queen" in memory_dirs else None,
)
@@ -771,9 +763,7 @@ async def run_unified_long_reflection(
if queen_memory_dir is not None and queen_id:
memory_dirs["queen"] = queen_memory_dir
manifest = _format_multi_scope_manifest(
memory_dirs, queen_id=queen_id if "queen" in memory_dirs else None
)
manifest = _format_multi_scope_manifest(memory_dirs, queen_id=queen_id if "queen" in memory_dirs else None)
user_msg = (
"## Current memory manifest across scopes\n\n"
f"{manifest}\n\n"
core/framework/config.py
+1 -3
View File
@@ -405,9 +405,7 @@ def _fetch_antigravity_credentials() -> tuple[str | None, str | None]:
import urllib.request
try:
req = urllib.request.Request(
_ANTIGRAVITY_CREDENTIALS_URL, headers={"User-Agent": "Hive/1.0"}
)
req = urllib.request.Request(_ANTIGRAVITY_CREDENTIALS_URL, headers={"User-Agent": "Hive/1.0"})
with urllib.request.urlopen(req, timeout=10) as resp:
content = resp.read().decode("utf-8")
id_match = re.search(r'ANTIGRAVITY_CLIENT_ID\s*=\s*"([^"]+)"', content)
core/framework/credentials/aden/client.py
+2 -6
View File
@@ -332,9 +332,7 @@ class AdenCredentialClient:
last_error = e
if attempt < self.config.retry_attempts - 1:
delay = self.config.retry_delay * (2**attempt)
logger.warning(
f"Aden request failed (attempt {attempt + 1}), retrying in {delay}s: {e}"
)
logger.warning(f"Aden request failed (attempt {attempt + 1}), retrying in {delay}s: {e}")
time.sleep(delay)
else:
raise AdenClientError(f"Failed to connect to Aden server: {e}") from e
@@ -347,9 +345,7 @@ class AdenCredentialClient:
):
raise
raise AdenClientError(
f"Request failed after {self.config.retry_attempts} attempts"
) from last_error
raise AdenClientError(f"Request failed after {self.config.retry_attempts} attempts") from last_error
def list_integrations(self) -> list[AdenIntegrationInfo]:
"""
core/framework/credentials/aden/provider.py
+2 -6
View File
@@ -192,9 +192,7 @@ class AdenSyncProvider(CredentialProvider):
f"Visit: {e.reauthorization_url or 'your Aden dashboard'}"
) from e
raise CredentialRefreshError(
f"Failed to refresh credential '{credential.id}': {e}"
) from e
raise CredentialRefreshError(f"Failed to refresh credential '{credential.id}': {e}") from e
except AdenClientError as e:
logger.error(f"Aden client error for '{credential.id}': {e}")
@@ -206,9 +204,7 @@ class AdenSyncProvider(CredentialProvider):
logger.warning(f"Aden unavailable, using cached token for '{credential.id}'")
return credential
raise CredentialRefreshError(
f"Aden server unavailable and token expired for '{credential.id}'"
) from e
raise CredentialRefreshError(f"Aden server unavailable and token expired for '{credential.id}'") from e
def validate(self, credential: CredentialObject) -> bool:
"""
core/framework/credentials/aden/storage.py
+1 -3
View File
@@ -168,9 +168,7 @@ class AdenCachedStorage(CredentialStorage):
if rid != credential_id:
result = self._load_by_id(rid)
if result is not None:
logger.info(
f"Loaded credential '{credential_id}' via provider index (id='{rid}')"
)
logger.info(f"Loaded credential '{credential_id}' via provider index (id='{rid}')")
return result
# Direct lookup (exact credential_id match)
core/framework/credentials/aden/tests/test_aden_sync.py
+2 -6
View File
@@ -493,9 +493,7 @@ class TestAdenCachedStorage:
assert loaded is not None
assert loaded.keys["access_token"].value.get_secret_value() == "cached-token"
def test_load_from_aden_when_stale(
self, cached_storage, local_storage, provider, mock_client, aden_response
):
def test_load_from_aden_when_stale(self, cached_storage, local_storage, provider, mock_client, aden_response):
"""Test load fetches from Aden when cache is stale."""
# Create stale cached credential
cred = CredentialObject(
@@ -521,9 +519,7 @@ class TestAdenCachedStorage:
assert loaded is not None
assert loaded.keys["access_token"].value.get_secret_value() == "test-access-token"
def test_load_falls_back_to_stale_when_aden_fails(
self, cached_storage, local_storage, provider, mock_client
):
def test_load_falls_back_to_stale_when_aden_fails(self, cached_storage, local_storage, provider, mock_client):
"""Test load falls back to stale cache when Aden fails."""
# Create stale cached credential
cred = CredentialObject(
core/framework/credentials/oauth2/base_provider.py
+3 -8
View File
@@ -95,9 +95,7 @@ class BaseOAuth2Provider(CredentialProvider):
self._client = httpx.Client(timeout=self.config.request_timeout)
except ImportError as e:
raise ImportError(
"OAuth2 provider requires 'httpx'. Install with: uv pip install httpx"
) from e
raise ImportError("OAuth2 provider requires 'httpx'. Install with: uv pip install httpx") from e
return self._client
def _close_client(self) -> None:
@@ -311,8 +309,7 @@ class BaseOAuth2Provider(CredentialProvider):
except OAuth2Error as e:
if e.error == "invalid_grant":
raise CredentialRefreshError(
f"Refresh token for '{credential.id}' is invalid or revoked. "
"Re-authorization required."
f"Refresh token for '{credential.id}' is invalid or revoked. Re-authorization required."
) from e
raise CredentialRefreshError(f"Failed to refresh '{credential.id}': {e}") from e
@@ -422,9 +419,7 @@ class BaseOAuth2Provider(CredentialProvider):
if response.status_code != 200 or "error" in response_data:
error = response_data.get("error", "unknown_error")
description = response_data.get("error_description", response.text)
raise OAuth2Error(
error=error, description=description, status_code=response.status_code
)
raise OAuth2Error(error=error, description=description, status_code=response.status_code)
return OAuth2Token.from_token_response(response_data)
core/framework/credentials/oauth2/lifecycle.py
+1 -3
View File
@@ -158,9 +158,7 @@ class TokenLifecycleManager:
"""
# Run in executor to avoid blocking
loop = asyncio.get_event_loop()
token = await loop.run_in_executor(
None, lambda: self.provider.client_credentials_grant(scopes=scopes)
)
token = await loop.run_in_executor(None, lambda: self.provider.client_credentials_grant(scopes=scopes))
self._save_token_to_store(token)
self._cached_token = token
core/framework/credentials/oauth2/zoho_provider.py
+1 -3
View File
@@ -100,9 +100,7 @@ class ZohoOAuth2Provider(BaseOAuth2Provider):
)
super().__init__(config, provider_id="zoho_crm_oauth2")
self._accounts_domain = base
self._api_domain = (
api_domain or os.getenv("ZOHO_API_DOMAIN", "https://www.zohoapis.com")
).rstrip("/")
self._api_domain = (api_domain or os.getenv("ZOHO_API_DOMAIN", "https://www.zohoapis.com")).rstrip("/")
@property
def supported_types(self) -> list[CredentialType]:
core/framework/credentials/setup.py
+2 -6
View File
@@ -268,9 +268,7 @@ class CredentialSetupSession:
self._print(f"{Colors.YELLOW}Initializing credential store...{Colors.NC}")
try:
generate_and_save_credential_key()
self._print(
f"{Colors.GREEN}✓ Encryption key saved to ~/.hive/secrets/credential_key{Colors.NC}"
)
self._print(f"{Colors.GREEN}✓ Encryption key saved to ~/.hive/secrets/credential_key{Colors.NC}")
return True
except Exception as e:
self._print(f"{Colors.RED}Failed to initialize credential store: {e}{Colors.NC}")
@@ -449,9 +447,7 @@ class CredentialSetupSession:
logger.warning("Unexpected error exporting credential to env", exc_info=True)
return True
else:
self._print(
f"{Colors.YELLOW}⚠ {cred.credential_name} not found in Aden account.{Colors.NC}"
)
self._print(f"{Colors.YELLOW}⚠ {cred.credential_name} not found in Aden account.{Colors.NC}")
self._print("Please connect this integration on https://hive.adenhq.com first.")
return False
except Exception as e:
core/framework/credentials/storage.py
+6 -15
View File
@@ -136,8 +136,7 @@ class EncryptedFileStorage(CredentialStorage):
from cryptography.fernet import Fernet
except ImportError as e:
raise ImportError(
"Encrypted storage requires 'cryptography'. "
"Install with: uv pip install cryptography"
"Encrypted storage requires 'cryptography'. Install with: uv pip install cryptography"
) from e
self.base_path = Path(base_path or self.DEFAULT_PATH).expanduser()
@@ -213,9 +212,7 @@ class EncryptedFileStorage(CredentialStorage):
json_bytes = self._fernet.decrypt(encrypted)
data = json.loads(json_bytes.decode("utf-8-sig"))
except Exception as e:
raise CredentialDecryptionError(
f"Failed to decrypt credential '{credential_id}': {e}"
) from e
raise CredentialDecryptionError(f"Failed to decrypt credential '{credential_id}': {e}") from e
# Deserialize
return self._deserialize_credential(data)
@@ -316,8 +313,7 @@ class EncryptedFileStorage(CredentialStorage):
visible_keys = [
name
for name in credential.keys.keys()
if name not in self.INDEX_INTERNAL_KEY_NAMES
and not name.startswith("_identity_")
if name not in self.INDEX_INTERNAL_KEY_NAMES and not name.startswith("_identity_")
]
# Earliest expiry across all keys (most likely the access_token).
@@ -336,9 +332,7 @@ class EncryptedFileStorage(CredentialStorage):
"key_names": sorted(visible_keys),
"created_at": credential.created_at.isoformat() if credential.created_at else None,
"updated_at": credential.updated_at.isoformat() if credential.updated_at else None,
"last_refreshed": (
credential.last_refreshed.isoformat() if credential.last_refreshed else None
),
"last_refreshed": (credential.last_refreshed.isoformat() if credential.last_refreshed else None),
"expires_at": earliest_expiry.isoformat() if earliest_expiry else None,
"auto_refresh": credential.auto_refresh,
"tags": list(credential.tags),
@@ -480,8 +474,7 @@ class EnvVarStorage(CredentialStorage):
def save(self, credential: CredentialObject) -> None:
"""Cannot save to environment variables at runtime."""
raise NotImplementedError(
"EnvVarStorage is read-only. Set environment variables "
"externally or use EncryptedFileStorage."
"EnvVarStorage is read-only. Set environment variables externally or use EncryptedFileStorage."
)
def load(self, credential_id: str) -> CredentialObject | None:
@@ -501,9 +494,7 @@ class EnvVarStorage(CredentialStorage):
def delete(self, credential_id: str) -> bool:
"""Cannot delete environment variables at runtime."""
raise NotImplementedError(
"EnvVarStorage is read-only. Unset environment variables externally."
)
raise NotImplementedError("EnvVarStorage is read-only. Unset environment variables externally.")
def list_all(self) -> list[str]:
"""List credentials that are available in environment."""
core/framework/credentials/store.py
+5 -15
View File
@@ -124,9 +124,7 @@ class CredentialStore:
"""
return self._providers.get(provider_id)
def get_provider_for_credential(
self, credential: CredentialObject
) -> CredentialProvider | None:
def get_provider_for_credential(self, credential: CredentialObject) -> CredentialProvider | None:
"""
Get the appropriate provider for a credential.
@@ -201,9 +199,7 @@ class CredentialStore:
cached = self._get_from_cache(credential_id)
if cached is not None:
if refresh_if_needed and self._should_refresh(cached):
return self._refresh_credential(
cached, raise_on_failure=raise_on_refresh_failure
)
return self._refresh_credential(cached, raise_on_failure=raise_on_refresh_failure)
return cached
# Load from storage
@@ -213,9 +209,7 @@ class CredentialStore:
# Refresh if needed
if refresh_if_needed and self._should_refresh(credential):
credential = self._refresh_credential(
credential, raise_on_failure=raise_on_refresh_failure
)
credential = self._refresh_credential(credential, raise_on_failure=raise_on_refresh_failure)
# Cache
self._add_to_cache(credential)
@@ -240,9 +234,7 @@ class CredentialStore:
Returns:
The key value or None if not found
"""
credential = self.get_credential(
credential_id, raise_on_refresh_failure=raise_on_refresh_failure
)
credential = self.get_credential(credential_id, raise_on_refresh_failure=raise_on_refresh_failure)
if credential is None:
return None
return credential.get_key(key_name)
@@ -266,9 +258,7 @@ class CredentialStore:
Returns:
The primary key value or None
"""
credential = self.get_credential(
credential_id, raise_on_refresh_failure=raise_on_refresh_failure
)
credential = self.get_credential(credential_id, raise_on_refresh_failure=raise_on_refresh_failure)
if credential is None:
return None
return credential.get_default_key()
core/framework/credentials/template.py
+2 -6
View File
@@ -88,9 +88,7 @@ class TemplateResolver:
if key_name:
value = credential.get_key(key_name)
if value is None:
raise CredentialKeyNotFoundError(
f"Key '{key_name}' not found in credential '{cred_id}'"
)
raise CredentialKeyNotFoundError(f"Key '{key_name}' not found in credential '{cred_id}'")
else:
# Use default key
value = credential.get_default_key()
@@ -126,9 +124,7 @@ class TemplateResolver:
... })
{"Authorization": "Bearer ghp_xxx", "X-API-Key": "BSAKxxx"}
"""
return {
key: self.resolve(value, fail_on_missing) for key, value in header_templates.items()
}
return {key: self.resolve(value, fail_on_missing) for key, value in header_templates.items()}
def resolve_params(
self,
core/framework/credentials/tests/test_credential_store.py
+9 -33
View File
@@ -130,9 +130,7 @@ class TestCredentialObject:
# With access_token
cred2 = CredentialObject(
id="test",
keys={
"access_token": CredentialKey(name="access_token", value=SecretStr("token-value"))
},
keys={"access_token": CredentialKey(name="access_token", value=SecretStr("token-value"))},
)
assert cred2.get_default_key() == "token-value"
@@ -297,9 +295,7 @@ class TestEncryptedFileStorage:
key = Fernet.generate_key().decode()
with patch.dict(os.environ, {"HIVE_CREDENTIAL_KEY": key}):
storage = EncryptedFileStorage(temp_dir)
cred = CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))}
)
cred = CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))})
storage.save(cred)
# Create new storage instance with same key
@@ -330,18 +326,10 @@ class TestCompositeStorage:
def test_read_from_primary(self):
"""Test reading from primary storage."""
primary = InMemoryStorage()
primary.save(
CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("primary"))}
)
)
primary.save(CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("primary"))}))
fallback = InMemoryStorage()
fallback.save(
CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("fallback"))}
)
)
fallback.save(CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("fallback"))}))
storage = CompositeStorage(primary, [fallback])
cred = storage.load("test")
@@ -353,11 +341,7 @@ class TestCompositeStorage:
"""Test fallback when credential not in primary."""
primary = InMemoryStorage()
fallback = InMemoryStorage()
fallback.save(
CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("fallback"))}
)
)
fallback.save(CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("fallback"))}))
storage = CompositeStorage(primary, [fallback])
cred = storage.load("test")
@@ -393,9 +377,7 @@ class TestStaticProvider:
def test_refresh_returns_unchanged(self):
"""Test that refresh returns credential unchanged."""
provider = StaticProvider()
cred = CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))}
)
cred = CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))})
refreshed = provider.refresh(cred)
assert refreshed.get_key("k") == "v"
@@ -403,9 +385,7 @@ class TestStaticProvider:
def test_validate_with_keys(self):
"""Test validation with keys present."""
provider = StaticProvider()
cred = CredentialObject(
id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))}
)
cred = CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))})
assert provider.validate(cred)
@@ -606,9 +586,7 @@ class TestCredentialStore:
storage = InMemoryStorage()
store = CredentialStore(storage=storage, cache_ttl_seconds=60)
storage.save(
CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))})
)
storage.save(CredentialObject(id="test", keys={"k": CredentialKey(name="k", value=SecretStr("v"))}))
# First load
store.get_credential("test")
@@ -686,9 +664,7 @@ class TestOAuth2Module:
from core.framework.credentials.oauth2 import OAuth2Config, TokenPlacement
# Valid config
config = OAuth2Config(
token_url="https://example.com/token", client_id="id", client_secret="secret"
)
config = OAuth2Config(token_url="https://example.com/token", client_id="id", client_secret="secret")
assert config.token_url == "https://example.com/token"
# Missing token_url
core/framework/credentials/validation.py
+6 -22
View File
@@ -160,15 +160,9 @@ class CredentialValidationResult:
if aden_nc:
if missing or invalid:
lines.append("")
lines.append(
"Aden integrations not connected "
"(ADEN_API_KEY is set but OAuth tokens unavailable):\n"
)
lines.append("Aden integrations not connected (ADEN_API_KEY is set but OAuth tokens unavailable):\n")
for c in aden_nc:
lines.append(
f" {c.env_var} for {_label(c)}"
f"\n Connect this integration at hive.adenhq.com first."
)
lines.append(f" {c.env_var} for {_label(c)}\n Connect this integration at hive.adenhq.com first.")
lines.append("\nIf you've already set up credentials, restart your terminal to load them.")
return "\n".join(lines)
@@ -270,8 +264,7 @@ def compute_unavailable_tools(nodes: list) -> tuple[set[str], list[str]]:
reason = "invalid"
messages.append(
f"{status.env_var} ({reason}) → drops {len(status.tools)} tool(s): "
f"{', '.join(status.tools[:6])}"
+ (f" +{len(status.tools) - 6} more" if len(status.tools) > 6 else "")
f"{', '.join(status.tools[:6])}" + (f" +{len(status.tools) - 6} more" if len(status.tools) > 6 else "")
)
return drop, messages
@@ -332,9 +325,7 @@ def validate_agent_credentials(
if os.environ.get("ADEN_API_KEY"):
_presync_aden_tokens(CREDENTIAL_SPECS, force=force_refresh)
env_mapping = {
(spec.credential_id or name): spec.env_var for name, spec in CREDENTIAL_SPECS.items()
}
env_mapping = {(spec.credential_id or name): spec.env_var for name, spec in CREDENTIAL_SPECS.items()}
env_storage = EnvVarStorage(env_mapping=env_mapping)
if os.environ.get("HIVE_CREDENTIAL_KEY"):
storage = CompositeStorage(primary=env_storage, fallbacks=[EncryptedFileStorage()])
@@ -368,12 +359,7 @@ def validate_agent_credentials(
available = store.is_available(cred_id)
# Aden-not-connected: ADEN_API_KEY set, Aden-only cred, but integration missing
is_aden_nc = (
not available
and has_aden_key
and spec.aden_supported
and not spec.direct_api_key_supported
)
is_aden_nc = not available and has_aden_key and spec.aden_supported and not spec.direct_api_key_supported
status = CredentialStatus(
credential_name=cred_name,
@@ -491,9 +477,7 @@ def validate_agent_credentials(
identity_data = result.details.get("identity")
if identity_data and isinstance(identity_data, dict):
try:
cred_obj = store.get_credential(
status.credential_id, refresh_if_needed=False
)
cred_obj = store.get_credential(status.credential_id, refresh_if_needed=False)
if cred_obj:
cred_obj.set_identity(**identity_data)
store.save_credential(cred_obj)
core/framework/host/agent_host.py
+22 -65
View File
@@ -205,9 +205,7 @@ class AgentHost:
DeprecationWarning,
stacklevel=2,
)
self._skills_manager = SkillsManager.from_precomputed(
skills_catalog_prompt, protocols_prompt
)
self._skills_manager = SkillsManager.from_precomputed(skills_catalog_prompt, protocols_prompt)
else:
# Bare constructor: auto-load defaults
self._skills_manager = SkillsManager()
@@ -248,9 +246,7 @@ class AgentHost:
self._tools = tools or []
self._tool_executor = tool_executor
self._accounts_prompt = accounts_prompt
self._dynamic_memory_provider_factory: Callable[[str], Callable[[], str] | None] | None = (
None
)
self._dynamic_memory_provider_factory: Callable[[str], Callable[[], str] | None] | None = None
self._accounts_data = accounts_data
self._tool_provider_map = tool_provider_map
@@ -419,8 +415,7 @@ class AgentHost:
event_types = [_ET(et) for et in tc.get("event_types", [])]
if not event_types:
logger.warning(
f"Entry point '{ep_id}' has trigger_type='event' "
"but no event_types in trigger_config"
f"Entry point '{ep_id}' has trigger_type='event' but no event_types in trigger_config"
)
continue
@@ -450,9 +445,7 @@ class AgentHost:
# Run in the same session as the primary entry
# point so memory (e.g. user-defined rules) is
# shared and logs land in one session directory.
session_state = self._get_primary_session_state(
exclude_entry_point=entry_point_id
)
session_state = self._get_primary_session_state(exclude_entry_point=entry_point_id)
exec_id = await self.trigger(
entry_point_id,
{"event": event.to_dict()},
@@ -505,8 +498,7 @@ class AgentHost:
from croniter import croniter
except ImportError as e:
raise RuntimeError(
"croniter is required for cron-based entry points. "
"Install it with: uv pip install croniter"
"croniter is required for cron-based entry points. Install it with: uv pip install croniter"
) from e
try:
@@ -548,9 +540,7 @@ class AgentHost:
"Cron '%s': paused, skipping tick",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + sleep_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + sleep_secs
await asyncio.sleep(max(0, sleep_secs))
continue
@@ -578,9 +568,7 @@ class AgentHost:
"Cron '%s': agent actively working, skipping tick",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + sleep_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + sleep_secs
await asyncio.sleep(max(0, sleep_secs))
continue
@@ -590,24 +578,18 @@ class AgentHost:
is_isolated = ep_spec and ep_spec.isolation_level == "isolated"
if is_isolated:
if _persistent_session_id:
session_state = {
"resume_session_id": _persistent_session_id
}
session_state = {"resume_session_id": _persistent_session_id}
else:
session_state = None
else:
session_state = self._get_primary_session_state(
exclude_entry_point=entry_point_id
)
session_state = self._get_primary_session_state(exclude_entry_point=entry_point_id)
# Gate: skip tick if no active session
if session_state is None:
logger.debug(
"Cron '%s': no active session, skipping",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + sleep_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + sleep_secs
await asyncio.sleep(max(0, sleep_secs))
continue
@@ -680,9 +662,7 @@ class AgentHost:
"Timer '%s': paused, skipping tick",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + interval_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + interval_secs
await asyncio.sleep(interval_secs)
continue
@@ -708,9 +688,7 @@ class AgentHost:
"Timer '%s': agent actively working, skipping tick",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + interval_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + interval_secs
await asyncio.sleep(interval_secs)
continue
@@ -720,24 +698,18 @@ class AgentHost:
is_isolated = ep_spec and ep_spec.isolation_level == "isolated"
if is_isolated:
if _persistent_session_id:
session_state = {
"resume_session_id": _persistent_session_id
}
session_state = {"resume_session_id": _persistent_session_id}
else:
session_state = None
else:
session_state = self._get_primary_session_state(
exclude_entry_point=entry_point_id
)
session_state = self._get_primary_session_state(exclude_entry_point=entry_point_id)
# Gate: skip tick if no active session
if session_state is None:
logger.debug(
"Timer '%s': no active session, skipping",
entry_point_id,
)
self._timer_next_fire[entry_point_id] = (
time.monotonic() + interval_secs
)
self._timer_next_fire[entry_point_id] = time.monotonic() + interval_secs
await asyncio.sleep(interval_secs)
continue
@@ -1152,8 +1124,7 @@ class AgentHost:
event_types = [_ET(et) for et in tc.get("event_types", [])]
if not event_types:
logger.warning(
"Entry point '%s::%s' has trigger_type='event' "
"but no event_types in trigger_config",
"Entry point '%s::%s' has trigger_type='event' but no event_types in trigger_config",
graph_id,
ep_id,
)
@@ -1301,24 +1272,18 @@ class AgentHost:
break
stream = reg.streams.get(local_ep)
if not stream:
logger.warning(
"Timer: no stream '%s' in '%s', stopping", local_ep, gid
)
logger.warning("Timer: no stream '%s' in '%s', stopping", local_ep, gid)
break
# Isolated entry points get their own session;
# shared ones join the primary session.
ep_spec = reg.entry_points.get(local_ep)
if ep_spec and ep_spec.isolation_level == "isolated":
if _persistent_session_id:
session_state = {
"resume_session_id": _persistent_session_id
}
session_state = {"resume_session_id": _persistent_session_id}
else:
session_state = None
else:
session_state = self._get_primary_session_state(
local_ep, source_graph_id=gid
)
session_state = self._get_primary_session_state(local_ep, source_graph_id=gid)
# Gate: skip tick if no active session
if session_state is None:
logger.debug(
@@ -1335,11 +1300,7 @@ class AgentHost:
session_state=session_state,
)
# Remember session ID for reuse on next tick
if (
not _persistent_session_id
and ep_spec
and ep_spec.isolation_level == "isolated"
):
if not _persistent_session_id and ep_spec and ep_spec.isolation_level == "isolated":
_persistent_session_id = exec_id
except Exception:
logger.error(
@@ -1597,9 +1558,7 @@ class AgentHost:
src_graph_id = source_graph_id or self._graph_id
src_reg = self._graphs.get(src_graph_id)
ep_spec = (
src_reg.entry_points.get(exclude_entry_point)
if src_reg
else self._entry_points.get(exclude_entry_point)
src_reg.entry_points.get(exclude_entry_point) if src_reg else self._entry_points.get(exclude_entry_point)
)
if ep_spec:
graph = src_reg.graph if src_reg else self.graph
@@ -1633,9 +1592,7 @@ class AgentHost:
# Filter to only input keys so stale outputs
# from previous triggers don't leak through.
if allowed_keys is not None:
buffer_data = {
k: v for k, v in full_buffer.items() if k in allowed_keys
}
buffer_data = {k: v for k, v in full_buffer.items() if k in allowed_keys}
else:
buffer_data = full_buffer
if buffer_data:
core/framework/host/colony_runtime.py
+11 -34
View File
@@ -15,7 +15,6 @@ import asyncio
import json
import logging
import time
import uuid
from collections import OrderedDict
from collections.abc import Callable
from dataclasses import dataclass, field
@@ -25,16 +24,13 @@ from typing import TYPE_CHECKING, Any
from framework.agent_loop.types import AgentContext, AgentSpec
from framework.host.event_bus import AgentEvent, EventBus, EventType
from framework.host.triggers import TriggerDefinition
from framework.host.worker import Worker, WorkerInfo, WorkerResult, WorkerStatus
from framework.observability import set_trace_context
from framework.host.worker import Worker, WorkerInfo, WorkerResult
from framework.schemas.goal import Goal
from framework.storage.concurrent import ConcurrentStorage
from framework.storage.session_store import SessionStore
if TYPE_CHECKING:
from framework.agent_loop.agent_loop import AgentLoop
from framework.llm.provider import LLMProvider, Tool
from framework.pipeline.runner import PipelineRunner
from framework.skills.manager import SkillsManagerConfig
from framework.tracker.runtime_log_store import RuntimeLogStore
@@ -195,9 +191,7 @@ class ColonyRuntime:
DeprecationWarning,
stacklevel=2,
)
self._skills_manager = SkillsManager.from_precomputed(
skills_catalog_prompt, protocols_prompt
)
self._skills_manager = SkillsManager.from_precomputed(skills_catalog_prompt, protocols_prompt)
else:
self._skills_manager = SkillsManager()
self._skills_manager.load()
@@ -210,9 +204,7 @@ class ColonyRuntime:
self._accounts_prompt = accounts_prompt
self._accounts_data = accounts_data
self._tool_provider_map = tool_provider_map
self._dynamic_memory_provider_factory: Callable[[str], Callable[[], str] | None] | None = (
None
)
self._dynamic_memory_provider_factory: Callable[[str], Callable[[], str] | None] | None = None
storage_path_obj = Path(storage_path) if isinstance(storage_path, str) else storage_path
self._storage_path: Path = storage_path_obj
@@ -560,9 +552,7 @@ class ColonyRuntime:
encoding="utf-8",
)
except (json.JSONDecodeError, OSError) as exc:
logger.warning(
"spawn fork: failed to copy queen meta.json: %s", exc
)
logger.warning("spawn fork: failed to copy queen meta.json: %s", exc)
# Append the task as the next user message so the worker's
# LLM sees it as the most recent turn in the conversation
@@ -674,9 +664,7 @@ class ColonyRuntime:
input_data=input_data,
)
worker_conv_store = FileConversationStore(
worker_storage / "conversations"
)
worker_conv_store = FileConversationStore(worker_storage / "conversations")
# AgentLoop takes bus/judge/config/executor at construction;
# LLM, tools, stream_id, execution_id all come from the
@@ -848,9 +836,7 @@ class ColonyRuntime:
if remaining <= 0:
break
try:
report = await asyncio.wait_for(
report_queue.get(), timeout=remaining
)
report = await asyncio.wait_for(report_queue.get(), timeout=remaining)
except TimeoutError:
break
wid = report.get("worker_id")
@@ -919,10 +905,7 @@ class ColonyRuntime:
return self._overseer
if not self._running:
raise RuntimeError(
"start_overseer requires the ColonyRuntime to be running "
"(call start() first)"
)
raise RuntimeError("start_overseer requires the ColonyRuntime to be running (call start() first)")
from framework.agent_loop.agent_loop import AgentLoop
from framework.storage.conversation_store import FileConversationStore
@@ -933,9 +916,7 @@ class ColonyRuntime:
# {colony_session}/conversations/. Workers get their own sub-dirs
# under workers/{worker_id}/; the overseer is the root occupant.
self._storage_path.mkdir(parents=True, exist_ok=True)
overseer_conv_store = FileConversationStore(
self._storage_path / "conversations"
)
overseer_conv_store = FileConversationStore(self._storage_path / "conversations")
agent_loop = AgentLoop(
event_bus=self._scoped_event_bus,
tool_executor=self._tool_executor,
@@ -1096,9 +1077,7 @@ class ColonyRuntime:
def get_worker_result(self, worker_id: str) -> WorkerResult | None:
return self._execution_results.get(worker_id)
async def wait_for_worker(
self, worker_id: str, timeout: float | None = None
) -> WorkerResult | None:
async def wait_for_worker(self, worker_id: str, timeout: float | None = None) -> WorkerResult | None:
worker = self._workers.get(worker_id)
if worker is None:
return self._execution_results.get(worker_id)
@@ -1106,7 +1085,7 @@ class ColonyRuntime:
return worker.info.result
try:
await asyncio.wait_for(asyncio.shield(worker._task_handle), timeout=timeout)
except asyncio.TimeoutError:
except TimeoutError:
return None
return worker.info.result
@@ -1147,9 +1126,7 @@ class ColonyRuntime:
if worker and worker.is_active:
loop = worker._agent_loop
if hasattr(loop, "inject_event"):
await loop.inject_event(
content, is_client_input=is_client_input, image_content=image_content
)
await loop.inject_event(content, is_client_input=is_client_input, image_content=image_content)
return True
return False
core/framework/host/event_bus.py
+1 -5
View File
@@ -446,11 +446,7 @@ class EventBus:
# iteration values. Without this, live SSE would use raw iterations
# while events.jsonl would use offset iterations, causing ID collisions
# on the frontend when replaying after cold resume.
if (
self._session_log_iteration_offset
and isinstance(event.data, dict)
and "iteration" in event.data
):
if self._session_log_iteration_offset and isinstance(event.data, dict) and "iteration" in event.data:
offset = self._session_log_iteration_offset
event.data = {**event.data, "iteration": event.data["iteration"] + offset}
core/framework/host/execution_manager.py
+3 -9
View File
@@ -452,9 +452,7 @@ class ExecutionManager:
for executor in self._active_executors.values():
node = executor.node_registry.get(node_id)
if node is not None and hasattr(node, "inject_event"):
await node.inject_event(
content, is_client_input=is_client_input, image_content=image_content
)
await node.inject_event(content, is_client_input=is_client_input, image_content=image_content)
return True
return False
@@ -669,9 +667,7 @@ class ExecutionManager:
if self._runtime_log_store:
from framework.tracker.runtime_logger import RuntimeLogger
runtime_logger = RuntimeLogger(
store=self._runtime_log_store, agent_id=self.graph.id
)
runtime_logger = RuntimeLogger(store=self._runtime_log_store, agent_id=self.graph.id)
# Derive storage from session_store (graph-specific for secondary
# graphs) so that all files — conversations, state, checkpoints,
@@ -887,9 +883,7 @@ class ExecutionManager:
if has_result and result.paused_at:
await self._write_session_state(execution_id, ctx, result=result)
else:
await self._write_session_state(
execution_id, ctx, error="Execution cancelled"
)
await self._write_session_state(execution_id, ctx, error="Execution cancelled")
# Emit SSE event so the frontend knows the execution stopped.
# The executor does NOT emit on CancelledError, so there is no
core/framework/host/shared_state.py
-2
View File
@@ -2,8 +2,6 @@
import asyncio
import logging
import time
from dataclasses import dataclass, field
from enum import StrEnum
from typing import Any
core/framework/host/stream_runtime.py
+2 -7
View File
@@ -136,9 +136,7 @@ class StreamDecisionTracker:
self._run_locks[execution_id] = asyncio.Lock()
self._current_nodes[execution_id] = "unknown"
logger.debug(
f"Started run {run_id} for execution {execution_id} in stream {self.stream_id}"
)
logger.debug(f"Started run {run_id} for execution {execution_id} in stream {self.stream_id}")
return run_id
def end_run(
@@ -334,10 +332,7 @@ class StreamDecisionTracker:
"""
run = self._runs.get(execution_id)
if run is None:
logger.warning(
f"report_problem called but no run for execution {execution_id}: "
f"[{severity}] {description}"
)
logger.warning(f"report_problem called but no run for execution {execution_id}: [{severity}] {description}")
return ""
return run.add_problem(
core/framework/host/webhook_server.py
+1 -2
View File
@@ -89,8 +89,7 @@ class WebhookServer:
)
await self._site.start()
logger.info(
f"Webhook server started on {self._config.host}:{self._config.port} "
f"with {len(self._routes)} route(s)"
f"Webhook server started on {self._config.host}:{self._config.port} with {len(self._routes)} route(s)"
)
async def stop(self) -> None:
core/framework/host/worker.py
+7 -24
View File
@@ -92,9 +92,7 @@ class Worker:
# result.json, data). Required when seed_conversation() is used —
# we deliberately do NOT fall back to CWD, which previously caused
# conversation parts to leak into the process working directory.
self._storage_path: Path | None = (
Path(storage_path) if storage_path is not None else None
)
self._storage_path: Path | None = Path(storage_path) if storage_path is not None else None
self._task_handle: asyncio.Task | None = None
self._started_at: float = 0.0
self._result: WorkerResult | None = None
@@ -153,14 +151,10 @@ class Worker:
if result.success:
self.status = WorkerStatus.COMPLETED
self._result = self._build_result(
result, duration, default_status="success"
)
self._result = self._build_result(result, duration, default_status="success")
else:
self.status = WorkerStatus.FAILED
self._result = self._build_result(
result, duration, default_status="failed"
)
self._result = self._build_result(result, duration, default_status="failed")
await self._emit_terminal_events(result)
@@ -292,11 +286,7 @@ class Worker:
# EXECUTION_COMPLETED / EXECUTION_FAILED (backwards-compat)
if agent_result is not None:
lifecycle_type = (
EventType.EXECUTION_COMPLETED
if agent_result.success
else EventType.EXECUTION_FAILED
)
lifecycle_type = EventType.EXECUTION_COMPLETED if agent_result.success else EventType.EXECUTION_FAILED
await self._event_bus.publish(
AgentEvent(
type=lifecycle_type,
@@ -309,11 +299,7 @@ class Worker:
"task": self.task,
"success": agent_result.success,
"error": agent_result.error,
"output_keys": (
list(agent_result.output.keys())
if agent_result.output
else []
),
"output_keys": (list(agent_result.output.keys()) if agent_result.output else []),
},
)
)
@@ -348,9 +334,7 @@ class Worker:
async def start_background(self) -> None:
"""Spawn the worker's run() as an asyncio background task."""
self._task_handle = asyncio.create_task(
self.run(), name=f"worker:{self.id}"
)
self._task_handle = asyncio.create_task(self.run(), name=f"worker:{self.id}")
# Surface any exception that escapes run(); without this callback
# a crash here only becomes visible when stop() eventually awaits
# the handle (and is silently lost if stop() is never called).
@@ -406,8 +390,7 @@ class Worker:
"""
if self.status != WorkerStatus.PENDING:
raise RuntimeError(
f"seed_conversation must be called before start_background "
f"(worker {self.id} is {self.status})"
f"seed_conversation must be called before start_background (worker {self.id} is {self.status})"
)
# Write parts directly to the worker's on-disk conversation store
core/framework/llm/anthropic.py
+1 -3
View File
@@ -50,9 +50,7 @@ class AnthropicProvider(LLMProvider):
# Delegate to LiteLLMProvider internally.
self.api_key = api_key or _get_api_key_from_credential_store()
if not self.api_key:
raise ValueError(
"Anthropic API key required. Set ANTHROPIC_API_KEY env var or pass api_key."
)
raise ValueError("Anthropic API key required. Set ANTHROPIC_API_KEY env var or pass api_key.")
self.model = model
core/framework/llm/antigravity.py
+8 -29
View File
@@ -53,17 +53,9 @@ _TOKEN_REFRESH_BUFFER_SECS = 60
# Credentials file in ~/.hive/ (native implementation)
_ACCOUNTS_FILE = Path.home() / ".hive" / "antigravity-accounts.json"
_IDE_STATE_DB_MAC = (
Path.home()
/ "Library"
/ "Application Support"
/ "Antigravity"
/ "User"
/ "globalStorage"
/ "state.vscdb"
)
_IDE_STATE_DB_LINUX = (
Path.home() / ".config" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
Path.home() / "Library" / "Application Support" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
)
_IDE_STATE_DB_LINUX = Path.home() / ".config" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
_IDE_STATE_DB_KEY = "antigravityUnifiedStateSync.oauthToken"
_BASE_HEADERS: dict[str, str] = {
@@ -368,9 +360,7 @@ def _to_gemini_contents(
def _map_finish_reason(reason: str) -> str:
return {"STOP": "stop", "MAX_TOKENS": "max_tokens", "OTHER": "tool_use"}.get(
(reason or "").upper(), "stop"
)
return {"STOP": "stop", "MAX_TOKENS": "max_tokens", "OTHER": "tool_use"}.get((reason or "").upper(), "stop")
def _parse_complete_response(raw: dict[str, Any], model: str) -> LLMResponse:
@@ -538,8 +528,7 @@ class AntigravityProvider(LLMProvider):
return self._access_token
raise RuntimeError(
"No valid Antigravity credentials. "
"Run: uv run python core/antigravity_auth.py auth account add"
"No valid Antigravity credentials. Run: uv run python core/antigravity_auth.py auth account add"
)
# --- Request building -------------------------------------------------- #
@@ -593,11 +582,7 @@ class AntigravityProvider(LLMProvider):
token = self._ensure_token()
body_bytes = json.dumps(body).encode("utf-8")
path = (
"/v1internal:streamGenerateContent?alt=sse"
if streaming
else "/v1internal:generateContent"
)
path = "/v1internal:streamGenerateContent?alt=sse" if streaming else "/v1internal:generateContent"
headers = {
**_BASE_HEADERS,
"Authorization": f"Bearer {token}",
@@ -619,9 +604,7 @@ class AntigravityProvider(LLMProvider):
if result:
self._access_token, self._token_expires_at = result
headers["Authorization"] = f"Bearer {self._access_token}"
req2 = urllib.request.Request(
url, data=body_bytes, headers=headers, method="POST"
)
req2 = urllib.request.Request(url, data=body_bytes, headers=headers, method="POST")
try:
return urllib.request.urlopen(req2, timeout=120) # noqa: S310
except urllib.error.HTTPError as exc2:
@@ -642,9 +625,7 @@ class AntigravityProvider(LLMProvider):
last_exc = exc
continue
raise RuntimeError(
f"All Antigravity endpoints failed. Last error: {last_exc}"
) from last_exc
raise RuntimeError(f"All Antigravity endpoints failed. Last error: {last_exc}") from last_exc
# --- LLMProvider interface --------------------------------------------- #
@@ -683,9 +664,7 @@ class AntigravityProvider(LLMProvider):
try:
body = self._build_body(messages, system, tools, max_tokens)
http_resp = self._post(body, streaming=True)
for event in _parse_sse_stream(
http_resp, self.model, self._thought_sigs.__setitem__
):
for event in _parse_sse_stream(http_resp, self.model, self._thought_sigs.__setitem__):
loop.call_soon_threadsafe(queue.put_nowait, event)
except Exception as exc:
logger.error("Antigravity stream error: %s", exc)
core/framework/llm/litellm.py
+21 -62
View File
@@ -100,9 +100,7 @@ def _patch_litellm_anthropic_oauth() -> None:
result["authorization"] = f"Bearer {token}"
# Merge the OAuth beta header with any existing beta headers.
existing_beta = result.get("anthropic-beta", "")
beta_parts = (
[b.strip() for b in existing_beta.split(",") if b.strip()] if existing_beta else []
)
beta_parts = [b.strip() for b in existing_beta.split(",") if b.strip()] if existing_beta else []
if ANTHROPIC_OAUTH_BETA_HEADER not in beta_parts:
beta_parts.append(ANTHROPIC_OAUTH_BETA_HEADER)
result["anthropic-beta"] = ",".join(beta_parts)
@@ -262,9 +260,7 @@ def _claude_code_billing_header(messages: list[dict[str, Any]]) -> str:
break
sampled = "".join(_sample_js_code_unit(first_text, i) for i in (4, 7, 20))
version_hash = hashlib.sha256(
f"{_CLAUDE_CODE_BILLING_SALT}{sampled}{CLAUDE_CODE_VERSION}".encode()
).hexdigest()
version_hash = hashlib.sha256(f"{_CLAUDE_CODE_BILLING_SALT}{sampled}{CLAUDE_CODE_VERSION}".encode()).hexdigest()
entrypoint = os.environ.get("CLAUDE_CODE_ENTRYPOINT", "").strip() or "cli"
return (
f"x-anthropic-billing-header: cc_version={CLAUDE_CODE_VERSION}.{version_hash[:3]}; "
@@ -336,9 +332,7 @@ def _prune_failed_request_dumps(max_files: int = MAX_FAILED_REQUEST_DUMPS) -> No
def _remember_openrouter_tool_compat_model(model: str) -> None:
"""Cache OpenRouter tool-compat fallback for a bounded time window."""
OPENROUTER_TOOL_COMPAT_MODEL_CACHE[model] = (
time.monotonic() + OPENROUTER_TOOL_COMPAT_CACHE_TTL_SECONDS
)
OPENROUTER_TOOL_COMPAT_MODEL_CACHE[model] = time.monotonic() + OPENROUTER_TOOL_COMPAT_CACHE_TTL_SECONDS
def _is_openrouter_tool_compat_cached(model: str) -> bool:
@@ -746,20 +740,14 @@ class LiteLLMProvider(LLMProvider):
eh.setdefault("user-agent", CLAUDE_CODE_USER_AGENT)
# The Codex ChatGPT backend (chatgpt.com/backend-api/codex) rejects
# several standard OpenAI params: max_output_tokens, stream_options.
self._codex_backend = bool(
self.api_base and "chatgpt.com/backend-api/codex" in self.api_base
)
self._codex_backend = bool(self.api_base and "chatgpt.com/backend-api/codex" in self.api_base)
# Antigravity routes through a local OpenAI-compatible proxy — no patches needed.
self._antigravity = bool(self.api_base and "localhost:8069" in self.api_base)
if litellm is None:
raise ImportError(
"LiteLLM is not installed. Please install it with: uv pip install litellm"
)
raise ImportError("LiteLLM is not installed. Please install it with: uv pip install litellm")
def reconfigure(
self, model: str, api_key: str | None = None, api_base: str | None = None
) -> None:
def reconfigure(self, model: str, api_key: str | None = None, api_base: str | None = None) -> None:
"""Hot-swap the model, API key, and/or base URL on this provider instance.
Since the same LiteLLMProvider object is shared by reference across the
@@ -784,9 +772,7 @@ class LiteLLMProvider(LLMProvider):
if self._claude_code_oauth:
eh = self.extra_kwargs.setdefault("extra_headers", {})
eh.setdefault("user-agent", CLAUDE_CODE_USER_AGENT)
self._codex_backend = bool(
self.api_base and "chatgpt.com/backend-api/codex" in self.api_base
)
self._codex_backend = bool(self.api_base and "chatgpt.com/backend-api/codex" in self.api_base)
self._antigravity = bool(self.api_base and "localhost:8069" in self.api_base)
# Note: The Codex ChatGPT backend is a Responses API endpoint at
@@ -809,9 +795,7 @@ class LiteLLMProvider(LLMProvider):
return HIVE_API_BASE
return None
def _completion_with_rate_limit_retry(
self, max_retries: int | None = None, **kwargs: Any
) -> Any:
def _completion_with_rate_limit_retry(self, max_retries: int | None = None, **kwargs: Any) -> Any:
"""Call litellm.completion with retry on 429 rate limit errors and empty responses.
When a :class:`KeyPool` is configured, rate-limited keys are rotated
@@ -843,15 +827,10 @@ class LiteLLMProvider(LLMProvider):
None,
)
if last_role == "assistant":
logger.debug(
"[retry] Empty response after assistant message — "
"expected, not retrying."
)
logger.debug("[retry] Empty response after assistant message — expected, not retrying.")
return response
finish_reason = (
response.choices[0].finish_reason if response.choices else "unknown"
)
finish_reason = response.choices[0].finish_reason if response.choices else "unknown"
# Dump full request to file for debugging
token_count, token_method = _estimate_tokens(model, messages)
dump_path = _dump_failed_request(
@@ -1050,9 +1029,7 @@ class LiteLLMProvider(LLMProvider):
# Async variants — non-blocking on the event loop
# ------------------------------------------------------------------
async def _acompletion_with_rate_limit_retry(
self, max_retries: int | None = None, **kwargs: Any
) -> Any:
async def _acompletion_with_rate_limit_retry(self, max_retries: int | None = None, **kwargs: Any) -> Any:
"""Async version of _completion_with_rate_limit_retry.
Uses litellm.acompletion and asyncio.sleep instead of blocking calls.
@@ -1078,15 +1055,10 @@ class LiteLLMProvider(LLMProvider):
None,
)
if last_role == "assistant":
logger.debug(
"[async-retry] Empty response after assistant message — "
"expected, not retrying."
)
logger.debug("[async-retry] Empty response after assistant message — expected, not retrying.")
return response
finish_reason = (
response.choices[0].finish_reason if response.choices else "unknown"
)
finish_reason = response.choices[0].finish_reason if response.choices else "unknown"
token_count, token_method = _estimate_tokens(model, messages)
dump_path = _dump_failed_request(
model=model,
@@ -1370,8 +1342,7 @@ class LiteLLMProvider(LLMProvider):
)
return text_tool_content, text_tool_calls
logger.info(
"[openrouter-tool-compat] %s returned non-JSON fallback content; "
"treating it as plain text.",
"[openrouter-tool-compat] %s returned non-JSON fallback content; treating it as plain text.",
self.model,
)
return content.strip(), []
@@ -1523,9 +1494,7 @@ class LiteLLMProvider(LLMProvider):
)
return repaired
raise ValueError(
f"Failed to parse tool call arguments for '{tool_name}' (likely truncated JSON)."
)
raise ValueError(f"Failed to parse tool call arguments for '{tool_name}' (likely truncated JSON).")
def _parse_openrouter_text_tool_calls(
self,
@@ -1682,11 +1651,7 @@ class LiteLLMProvider(LLMProvider):
return [
message
for message in full_messages
if not (
message.get("role") == "assistant"
and not message.get("content")
and not message.get("tool_calls")
)
if not (message.get("role") == "assistant" and not message.get("content") and not message.get("tool_calls"))
]
async def _acomplete_via_openrouter_tool_compat(
@@ -1914,8 +1879,8 @@ class LiteLLMProvider(LLMProvider):
if logger.isEnabledFor(logging.DEBUG) and full_messages:
import json as _json
from pathlib import Path as _Path
from datetime import datetime as _dt
from pathlib import Path as _Path
_debug_dir = _Path.home() / ".hive" / "debug_logs"
_debug_dir.mkdir(parents=True, exist_ok=True)
@@ -1939,9 +1904,7 @@ class LiteLLMProvider(LLMProvider):
}
)
try:
_dump_file.write_text(
_json.dumps(_summary, indent=2, ensure_ascii=False), encoding="utf-8"
)
_dump_file.write_text(_json.dumps(_summary, indent=2, ensure_ascii=False), encoding="utf-8")
logger.debug("[LLM-MSG] %d messages dumped to %s", len(full_messages), _dump_file)
except Exception:
pass
@@ -1966,9 +1929,7 @@ class LiteLLMProvider(LLMProvider):
full_messages = [
m
for m in full_messages
if not (
m.get("role") == "assistant" and not m.get("content") and not m.get("tool_calls")
)
if not (m.get("role") == "assistant" and not m.get("content") and not m.get("tool_calls"))
]
kwargs: dict[str, Any] = {
@@ -2154,8 +2115,7 @@ class LiteLLMProvider(LLMProvider):
else getattr(usage, "cache_read_input_tokens", 0) or 0
)
logger.debug(
"[tokens] finish-chunk usage: "
"input=%d output=%d cached=%d model=%s",
"[tokens] finish-chunk usage: input=%d output=%d cached=%d model=%s",
input_tokens,
output_tokens,
cached_tokens,
@@ -2202,8 +2162,7 @@ class LiteLLMProvider(LLMProvider):
else getattr(_usage, "cache_read_input_tokens", 0) or 0
)
logger.debug(
"[tokens] post-loop chunks fallback:"
" input=%d output=%d cached=%d model=%s",
"[tokens] post-loop chunks fallback: input=%d output=%d cached=%d model=%s",
input_tokens,
output_tokens,
cached_tokens,
core/framework/llm/model_catalog.py
+7 -23
View File
@@ -50,9 +50,7 @@ def _validate_model_catalog(data: dict[str, Any]) -> dict[str, Any]:
if not isinstance(model_id, str) or not model_id.strip():
raise ModelCatalogError(f"{model_path}.id must be a non-empty string")
if model_id in seen_model_ids:
raise ModelCatalogError(
f"Duplicate model id {model_id!r} in {provider_path}.models"
)
raise ModelCatalogError(f"Duplicate model id {model_id!r} in {provider_path}.models")
seen_model_ids.add(model_id)
if model_id == default_model:
@@ -91,17 +89,11 @@ def _validate_model_catalog(data: dict[str, Any]) -> dict[str, Any]:
api_base = preset_map.get("api_base")
if api_base is not None and (not isinstance(api_base, str) or not api_base.strip()):
raise ModelCatalogError(
f"{preset_path}.api_base must be a non-empty string when present"
)
raise ModelCatalogError(f"{preset_path}.api_base must be a non-empty string when present")
api_key_env_var = preset_map.get("api_key_env_var")
if api_key_env_var is not None and (
not isinstance(api_key_env_var, str) or not api_key_env_var.strip()
):
raise ModelCatalogError(
f"{preset_path}.api_key_env_var must be a non-empty string when present"
)
if api_key_env_var is not None and (not isinstance(api_key_env_var, str) or not api_key_env_var.strip()):
raise ModelCatalogError(f"{preset_path}.api_key_env_var must be a non-empty string when present")
for key in ("max_tokens", "max_context_tokens"):
value = preset_map.get(key)
@@ -110,9 +102,7 @@ def _validate_model_catalog(data: dict[str, Any]) -> dict[str, Any]:
model_choices = preset_map.get("model_choices")
if model_choices is not None:
for idx, choice in enumerate(
_require_list(model_choices, f"{preset_path}.model_choices")
):
for idx, choice in enumerate(_require_list(model_choices, f"{preset_path}.model_choices")):
choice_path = f"{preset_path}.model_choices[{idx}]"
choice_map = _require_mapping(choice, choice_path)
choice_id = choice_map.get("id")
@@ -144,19 +134,13 @@ def load_model_catalog() -> dict[str, Any]:
def get_models_catalogue() -> dict[str, list[dict[str, Any]]]:
"""Return provider -> model list."""
providers = load_model_catalog()["providers"]
return {
provider_id: copy.deepcopy(provider_info["models"])
for provider_id, provider_info in providers.items()
}
return {provider_id: copy.deepcopy(provider_info["models"]) for provider_id, provider_info in providers.items()}
def get_default_models() -> dict[str, str]:
"""Return provider -> default model id."""
providers = load_model_catalog()["providers"]
return {
provider_id: str(provider_info["default_model"])
for provider_id, provider_info in providers.items()
}
return {provider_id: str(provider_info["default_model"]) for provider_id, provider_info in providers.items()}
def get_provider_models(provider: str) -> list[dict[str, Any]]:
core/framework/loader/agent_loader.py
+10 -33
View File
@@ -9,7 +9,7 @@ from datetime import UTC
from pathlib import Path
from typing import Any
from framework.config import get_hive_config, get_max_context_tokens, get_preferred_model
from framework.config import get_hive_config, get_preferred_model
from framework.credentials.validation import (
ensure_credential_key_env as _ensure_credential_key_env,
)
@@ -20,14 +20,12 @@ from framework.loader.preload_validation import run_preload_validation
from framework.loader.tool_registry import ToolRegistry
from framework.orchestrator import Goal
from framework.orchestrator.edge import (
DEFAULT_MAX_TOKENS,
EdgeCondition,
EdgeSpec,
GraphSpec,
)
from framework.orchestrator.node import NodeSpec
from framework.orchestrator.orchestrator import ExecutionResult
from framework.tools.flowchart_utils import generate_fallback_flowchart
logger = logging.getLogger(__name__)
@@ -555,18 +553,10 @@ def get_kimi_code_token() -> str | None:
# VSCode-style SQLite state database under the key
# "antigravityUnifiedStateSync.oauthToken" as a base64-encoded protobuf blob.
ANTIGRAVITY_IDE_STATE_DB = (
Path.home()
/ "Library"
/ "Application Support"
/ "Antigravity"
/ "User"
/ "globalStorage"
/ "state.vscdb"
Path.home() / "Library" / "Application Support" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
)
# Linux fallback for the IDE state DB
ANTIGRAVITY_IDE_STATE_DB_LINUX = (
Path.home() / ".config" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
)
ANTIGRAVITY_IDE_STATE_DB_LINUX = Path.home() / ".config" / "Antigravity" / "User" / "globalStorage" / "state.vscdb"
# Antigravity credentials stored by native OAuth implementation
ANTIGRAVITY_AUTH_FILE = Path.home() / ".hive" / "antigravity-accounts.json"
@@ -710,9 +700,7 @@ def _is_antigravity_token_expired(auth_data: dict) -> bool:
return True
elif isinstance(last_refresh_val, str):
try:
last_refresh_val = datetime.fromisoformat(
last_refresh_val.replace("Z", "+00:00")
).timestamp()
last_refresh_val = datetime.fromisoformat(last_refresh_val.replace("Z", "+00:00")).timestamp()
except (ValueError, TypeError):
return True
@@ -843,8 +831,7 @@ def get_antigravity_token() -> str | None:
return token_data["access_token"]
logger.warning(
"Antigravity token refresh failed. "
"Re-open the Antigravity IDE or run 'antigravity-auth accounts add'."
"Antigravity token refresh failed. Re-open the Antigravity IDE or run 'antigravity-auth accounts add'."
)
return access_token
@@ -1297,11 +1284,7 @@ class AgentLoader:
# Evict cached submodules first (e.g. deep_research_agent.nodes,
# deep_research_agent.agent) so the top-level reload picks up
# changes in the entire package — not just __init__.py.
stale = [
name
for name in sys.modules
if name == package_name or name.startswith(f"{package_name}.")
]
stale = [name for name in sys.modules if name == package_name or name.startswith(f"{package_name}.")]
for name in stale:
del sys.modules[name]
@@ -1350,7 +1333,7 @@ class AgentLoader:
if not worker_jsons:
raise FileNotFoundError(f"No worker config found in {agent_path}")
from framework.orchestrator.edge import EdgeSpec, GraphSpec
from framework.orchestrator.edge import GraphSpec
from framework.orchestrator.goal import Constraint, Goal as GoalModel, SuccessCriterion
from framework.orchestrator.node import NodeSpec
@@ -1555,7 +1538,6 @@ class AgentLoader:
]
# Merge user-configured stages from ~/.hive/configuration.json
from framework.config import get_hive_config
from framework.pipeline.registry import build_pipeline_from_config
hive_config = get_hive_config()
@@ -1568,9 +1550,7 @@ class AgentLoader:
if agent_json.exists():
try:
agent_pipeline = (
_json.loads(agent_json.read_text(encoding="utf-8"))
.get("pipeline", {})
.get("stages", [])
_json.loads(agent_json.read_text(encoding="utf-8")).get("pipeline", {}).get("stages", [])
)
if agent_pipeline:
agent_stages = build_pipeline_from_config(agent_pipeline)
@@ -1986,8 +1966,7 @@ class AgentLoader:
for sc in self.goal.success_criteria
],
constraints=[
{"id": c.id, "description": c.description, "type": c.constraint_type}
for c in self.goal.constraints
{"id": c.id, "description": c.description, "type": c.constraint_type} for c in self.goal.constraints
],
required_tools=sorted(required_tools),
has_tools_module=(self.agent_path / "tools.py").exists(),
@@ -2058,9 +2037,7 @@ class AgentLoader:
if api_key_env and not os.environ.get(api_key_env):
if api_key_env not in missing_credentials:
missing_credentials.append(api_key_env)
warnings.append(
f"Agent has LLM nodes but {api_key_env} not set (model: {self.model})"
)
warnings.append(f"Agent has LLM nodes but {api_key_env} not set (model: {self.model})")
return ValidationResult(
valid=len(errors) == 0,
core/framework/loader/cli.py
+25 -37
View File
@@ -25,7 +25,6 @@ from pathlib import Path
from typing import Any
from urllib import error as urlerror, parse as urlparse, request as urlrequest
# ---------------------------------------------------------------------------
# Public registration
# ---------------------------------------------------------------------------
@@ -127,10 +126,7 @@ def cmd_serve(args: argparse.Namespace) -> int:
def _request_shutdown(signame: str) -> None:
signal_count["n"] += 1
if signal_count["n"] == 1:
print(
f"\nReceived {signame}, shutting down gracefully… "
"(press Ctrl+C again to force quit)"
)
print(f"\nReceived {signame}, shutting down gracefully… (press Ctrl+C again to force quit)")
shutdown_event.set()
else:
# Second Ctrl+C (or SIGTERM) — the user is done waiting.
@@ -171,9 +167,7 @@ def cmd_serve(args: argparse.Namespace) -> int:
print(f"Colony not found: {colony_arg}")
continue
try:
session = await manager.create_session_with_worker_colony(
str(colony_path), model=model
)
session = await manager.create_session_with_worker_colony(str(colony_path), model=model)
info = session.worker_info
name = info.name if info else session.colony_id
print(f"Loaded colony: {session.colony_id} ({name}) → session {session.id}")
@@ -319,12 +313,14 @@ def cmd_queen_sessions(args: argparse.Namespace) -> int:
meta = json.loads(meta_path.read_text(encoding="utf-8"))
except Exception:
meta = {}
rows.append({
"session_id": session_dir.name,
"phase": meta.get("phase", "?"),
"agent_path": meta.get("agent_path", ""),
"colony_fork": bool(meta.get("colony_fork")),
})
rows.append(
{
"session_id": session_dir.name,
"phase": meta.get("phase", "?"),
"agent_path": meta.get("agent_path", ""),
"colony_fork": bool(meta.get("colony_fork")),
}
)
if args.json:
print(json.dumps(rows, indent=2))
@@ -398,18 +394,18 @@ def cmd_colony_list(args: argparse.Namespace) -> int:
except Exception:
meta = {}
worker_count = sum(
1
for f in path.iterdir()
if f.is_file() and f.suffix == ".json" and f.stem not in _RESERVED_JSON_STEMS
1 for f in path.iterdir() if f.is_file() and f.suffix == ".json" and f.stem not in _RESERVED_JSON_STEMS
)
rows.append(
{
"name": path.name,
"queen_name": meta.get("queen_name", ""),
"queen_session_id": meta.get("queen_session_id", ""),
"workers": worker_count,
"created_at": meta.get("created_at", ""),
"path": str(path),
}
)
rows.append({
"name": path.name,
"queen_name": meta.get("queen_name", ""),
"queen_session_id": meta.get("queen_session_id", ""),
"workers": worker_count,
"created_at": meta.get("created_at", ""),
"path": str(path),
})
if args.json:
print(json.dumps(rows, indent=2))
@@ -422,9 +418,7 @@ def cmd_colony_list(args: argparse.Namespace) -> int:
print(f"{'NAME':<24} {'QUEEN':<28} {'WORKERS':<8} CREATED")
print("-" * 90)
for r in rows:
print(
f"{r['name']:<24} {r['queen_name']:<28} {r['workers']:<8} {r['created_at'][:19]}"
)
print(f"{r['name']:<24} {r['queen_name']:<28} {r['workers']:<8} {r['created_at'][:19]}")
return 0
@@ -651,9 +645,7 @@ def _http_get(url: str, timeout: float = 10.0) -> dict:
def _http_post(url: str, body: dict, timeout: float = 30.0) -> dict:
data = json.dumps(body).encode("utf-8")
req = urlrequest.Request(
url, data=data, method="POST", headers={"Content-Type": "application/json"}
)
req = urlrequest.Request(url, data=data, method="POST", headers={"Content-Type": "application/json"})
with urlrequest.urlopen(req, timeout=timeout) as r:
return json.loads(r.read().decode("utf-8"))
@@ -709,9 +701,7 @@ def _open_browser(url: str) -> None:
try:
if sys.platform == "darwin":
subprocess.Popen(
["open", url], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
)
subprocess.Popen(["open", url], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
elif sys.platform == "win32":
subprocess.Popen(
["cmd", "/c", "start", "", url],
@@ -719,9 +709,7 @@ def _open_browser(url: str) -> None:
stderr=subprocess.DEVNULL,
)
elif sys.platform == "linux":
subprocess.Popen(
["xdg-open", url], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
)
subprocess.Popen(["xdg-open", url], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
except Exception:
pass
core/framework/loader/mcp_client.py
+7 -18
View File
@@ -267,9 +267,7 @@ class MCPClient:
try:
response = self._http_client.get("/health")
response.raise_for_status()
logger.info(
f"Connected to MCP server '{self.config.name}' via HTTP at {self.config.url}"
)
logger.info(f"Connected to MCP server '{self.config.name}' via HTTP at {self.config.url}")
except Exception as e:
logger.warning(f"Health check failed for MCP server '{self.config.name}': {e}")
# Continue anyway, server might not have health endpoint
@@ -377,12 +375,8 @@ class MCPClient:
self._tools[tool.name] = tool
tool_names = list(self._tools.keys())
logger.info(
f"Discovered {len(self._tools)} tools from '{self.config.name}'"
)
logger.debug(
f"Discovered tools from '{self.config.name}': {tool_names}"
)
logger.info(f"Discovered {len(self._tools)} tools from '{self.config.name}'")
logger.debug(f"Discovered tools from '{self.config.name}': {tool_names}")
except Exception as e:
logger.error(f"Failed to discover tools from '{self.config.name}': {e}")
raise
@@ -467,6 +461,7 @@ class MCPClient:
)
if self.config.transport == "stdio":
def _stdio_call() -> Any:
with self._stdio_call_lock:
return self._run_async(self._call_tool_stdio_async(tool_name, arguments))
@@ -669,9 +664,7 @@ class MCPClient:
if self._session:
await self._session.__aexit__(None, None, None)
except asyncio.CancelledError:
logger.warning(
"MCP session cleanup was cancelled; proceeding with best-effort shutdown"
)
logger.warning("MCP session cleanup was cancelled; proceeding with best-effort shutdown")
except Exception as e:
logger.warning(f"Error closing MCP session: {e}")
finally:
@@ -682,9 +675,7 @@ class MCPClient:
if self._stdio_context:
await self._stdio_context.__aexit__(None, None, None)
except asyncio.CancelledError:
logger.debug(
"STDIO context cleanup was cancelled; proceeding with best-effort shutdown"
)
logger.debug("STDIO context cleanup was cancelled; proceeding with best-effort shutdown")
except Exception as e:
msg = str(e).lower()
if "cancel scope" in msg or "different task" in msg:
@@ -725,9 +716,7 @@ class MCPClient:
# any exceptions that may occur if the loop stops between these calls.
if self._loop.is_running():
try:
cleanup_future = asyncio.run_coroutine_threadsafe(
self._cleanup_stdio_async(), self._loop
)
cleanup_future = asyncio.run_coroutine_threadsafe(self._cleanup_stdio_async(), self._loop)
cleanup_future.result(timeout=self._CLEANUP_TIMEOUT)
cleanup_attempted = True
except TimeoutError:
core/framework/loader/mcp_connection_manager.py
+5 -13
View File
@@ -74,8 +74,7 @@ class MCPConnectionManager:
if not should_connect:
if not transition_event.wait(timeout=_TRANSITION_TIMEOUT):
logger.warning(
"Timed out waiting for transition on MCP server '%s', "
"forcing cleanup and retrying",
"Timed out waiting for transition on MCP server '%s', forcing cleanup and retrying",
server_name,
)
with self._pool_lock:
@@ -99,10 +98,7 @@ class MCPConnectionManager:
current = self._transitions.get(server_name)
if current is transition_event:
self._transitions.pop(server_name, None)
if (
server_name not in self._pool
and self._refcounts.get(server_name, 0) <= 0
):
if server_name not in self._pool and self._refcounts.get(server_name, 0) <= 0:
self._configs.pop(server_name, None)
transition_event.set()
raise
@@ -324,8 +320,7 @@ class MCPConnectionManager:
self._transitions.pop(server_name, None)
transition_event.set()
logger.info(
"Reconnected MCP server '%s' but refcount dropped to 0, "
"discarding new client",
"Reconnected MCP server '%s' but refcount dropped to 0, discarding new client",
server_name,
)
try:
@@ -336,9 +331,7 @@ class MCPConnectionManager:
server_name,
exc_info=True,
)
raise KeyError(
f"MCP server '{server_name}' was fully released during reconnect"
)
raise KeyError(f"MCP server '{server_name}' was fully released during reconnect")
self._pool[server_name] = new_client
self._configs[server_name] = config
@@ -380,8 +373,7 @@ class MCPConnectionManager:
all_resolved = all(event.wait(timeout=_TRANSITION_TIMEOUT) for event in pending)
if not all_resolved:
logger.warning(
"Timed out waiting for pending transitions during cleanup, "
"forcing cleanup of stuck transitions",
"Timed out waiting for pending transitions during cleanup, forcing cleanup of stuck transitions",
)
with self._pool_lock:
for sn, evt in list(self._transitions.items()):
core/framework/loader/mcp_errors.py
+1 -3
View File
@@ -23,9 +23,7 @@ class MCPError(ValueError):
self.what = what
self.why = why
self.fix = fix
self.message = (
f"[{self.code.value}]\nWhat failed: {self.what}\nWhy: {self.why}\nFix: {self.fix}"
)
self.message = f"[{self.code.value}]\nWhat failed: {self.what}\nWhy: {self.why}\nFix: {self.fix}"
super().__init__(self.message)
core/framework/loader/mcp_registry.py
+3 -8
View File
@@ -24,9 +24,7 @@ from framework.loader.mcp_errors import (
logger = logging.getLogger(__name__)
DEFAULT_INDEX_URL = (
"https://raw.githubusercontent.com/aden-hive/hive-mcp-registry/main/registry_index.json"
)
DEFAULT_INDEX_URL = "https://raw.githubusercontent.com/aden-hive/hive-mcp-registry/main/registry_index.json"
DEFAULT_REFRESH_INTERVAL_HOURS = 24
_LAST_FETCHED_FILENAME = "last_fetched"
_LEGACY_LAST_FETCHED_FILENAME = "last_fetched.json"
@@ -140,9 +138,7 @@ class MCPRegistry:
)
added.append(name)
except MCPError as exc:
logger.warning(
"MCPRegistry.ensure_defaults: failed to seed '%s': %s", name, exc
)
logger.warning("MCPRegistry.ensure_defaults: failed to seed '%s': %s", name, exc)
if added:
logger.info("MCPRegistry: seeded default local servers: %s", added)
@@ -709,8 +705,7 @@ class MCPRegistry:
pinned_version = versions[name]
if installed_version != pinned_version:
logger.warning(
"Server '%s' version mismatch: installed=%s, pinned=%s. "
"Run: hive mcp update %s",
"Server '%s' version mismatch: installed=%s, pinned=%s. Run: hive mcp update %s",
name,
installed_version,
pinned_version,
core/framework/loader/mcp_registry_cli.py
+11 -30
View File
@@ -151,10 +151,7 @@ def _parse_key_value_pairs(values: list[str]) -> dict[str, str]:
result = {}
for item in values:
if "=" not in item:
raise ValueError(
f"Invalid format: '{item}'. Expected KEY=VALUE.\n"
f"Example: --set JIRA_API_TOKEN=abc123"
)
raise ValueError(f"Invalid format: '{item}'. Expected KEY=VALUE.\nExample: --set JIRA_API_TOKEN=abc123")
key, _, value = item.partition("=")
if not key:
raise ValueError(f"Invalid format: '{item}'. Key cannot be empty.")
@@ -300,12 +297,8 @@ def register_mcp_commands(subparsers) -> None:
# ── install ──
install_p = mcp_sub.add_parser("install", help="Install a server from the registry")
install_p.add_argument("name", help="Server name in the registry")
install_p.add_argument(
"--version", dest="version", default=None, help="Pin to a specific version"
)
install_p.add_argument(
"--transport", default=None, help="Override default transport (stdio, http, unix, sse)"
)
install_p.add_argument("--version", dest="version", default=None, help="Pin to a specific version")
install_p.add_argument("--transport", default=None, help="Override default transport (stdio, http, unix, sse)")
install_p.set_defaults(func=cmd_mcp_install)
# ── add ──
@@ -342,9 +335,7 @@ def register_mcp_commands(subparsers) -> None:
# ── list ──
list_p = mcp_sub.add_parser("list", help="List servers")
list_p.add_argument(
"--available", action="store_true", help="Show available servers from registry"
)
list_p.add_argument("--available", action="store_true", help="Show available servers from registry")
list_p.add_argument("--json", dest="output_json", action="store_true", help="Output as JSON")
list_p.set_defaults(func=cmd_mcp_list)
@@ -364,9 +355,7 @@ def register_mcp_commands(subparsers) -> None:
metavar="KEY=VAL",
help="Set environment variable overrides",
)
config_p.add_argument(
"--set-header", dest="set_header", nargs="+", metavar="KEY=VAL", help="Set header overrides"
)
config_p.add_argument("--set-header", dest="set_header", nargs="+", metavar="KEY=VAL", help="Set header overrides")
config_p.set_defaults(func=cmd_mcp_config)
# ── search ──
@@ -389,9 +378,7 @@ def register_mcp_commands(subparsers) -> None:
init_p.set_defaults(func=cmd_mcp_init)
# ── update ──
update_p = mcp_sub.add_parser(
"update", help="Update installed servers or refresh the registry index"
)
update_p = mcp_sub.add_parser("update", help="Update installed servers or refresh the registry index")
update_p.add_argument(
"name",
nargs="?",
@@ -495,8 +482,7 @@ def _cmd_mcp_add_from_manifest(registry, manifest_path: str) -> int:
manifest = json.loads(path.read_text(encoding="utf-8"))
except json.JSONDecodeError as exc:
print(
f"Error: invalid JSON in {manifest_path}: {exc}\n"
f"Validate with: python -m json.tool {manifest_path}",
f"Error: invalid JSON in {manifest_path}: {exc}\nValidate with: python -m json.tool {manifest_path}",
file=sys.stderr,
)
return 1
@@ -695,8 +681,7 @@ def cmd_mcp_config(args) -> int:
server = registry.get_server(args.name)
if server is None:
print(
f"Error: server '{args.name}' is not installed.\n"
f"Run 'hive mcp list' to see installed servers.",
f"Error: server '{args.name}' is not installed.\nRun 'hive mcp list' to see installed servers.",
file=sys.stderr,
)
return 1
@@ -822,8 +807,7 @@ def cmd_mcp_update(args) -> int:
count = registry.update_index()
except Exception as exc:
print(
f"Error: failed to update registry index: {exc}\n"
f"Check your network connection and try again.",
f"Error: failed to update registry index: {exc}\nCheck your network connection and try again.",
file=sys.stderr,
)
return 1
@@ -832,9 +816,7 @@ def cmd_mcp_update(args) -> int:
# Step 2: update all installed registry servers (skip local/pinned)
installed = registry.list_installed()
registry_servers = [
s for s in installed if s.get("source") == "registry" and not s.get("pinned")
]
registry_servers = [s for s in installed if s.get("source") == "registry" and not s.get("pinned")]
if not registry_servers:
return 0
@@ -862,8 +844,7 @@ def _cmd_mcp_update_server(name: str, registry=None) -> int:
server = registry.get_server(name)
if server is None:
print(
f"Error: server '{name}' is not installed.\n"
f"Run 'hive mcp install {name}' to install it.",
f"Error: server '{name}' is not installed.\nRun 'hive mcp install {name}' to install it.",
file=sys.stderr,
)
return 1
core/framework/loader/preload_validation.py
+1 -3
View File
@@ -98,9 +98,7 @@ def validate_credentials(
if not result.success:
# Preserve the original validation_result so callers can
# inspect which credentials are still missing.
exc = CredentialError(
"Credential setup incomplete. Run again after configuring the required credentials."
)
exc = CredentialError("Credential setup incomplete. Run again after configuring the required credentials.")
if hasattr(e, "validation_result"):
exc.validation_result = e.validation_result # type: ignore[attr-defined]
if hasattr(e, "failed_cred_names"):
core/framework/loader/tool_registry.py
+8 -28
View File
@@ -257,10 +257,7 @@ class ToolRegistry:
str(e),
)
return {
"error": (
f"Invalid JSON response from tool '{tool_name}': "
f"{str(e)}"
),
"error": (f"Invalid JSON response from tool '{tool_name}': {str(e)}"),
"raw_content": result.content,
}
return result
@@ -435,9 +432,7 @@ class ToolRegistry:
registry = ToolRegistry()
return registry._resolve_mcp_server_config(server_config, base_dir)
def _resolve_mcp_server_config(
self, server_config: dict[str, Any], base_dir: Path
) -> dict[str, Any]:
def _resolve_mcp_server_config(self, server_config: dict[str, Any], base_dir: Path) -> dict[str, Any]:
"""Resolve cwd and script paths for MCP stdio servers (Windows compatibility).
On Windows, passing cwd to subprocess can cause WinError 267. We use cwd=None
@@ -552,8 +547,7 @@ class ToolRegistry:
server_list = [{"name": name, **cfg} for name, cfg in config.items()]
resolved_server_list = [
self._resolve_mcp_server_config(server_config, base_dir)
for server_config in server_list
self._resolve_mcp_server_config(server_config, base_dir) for server_config in server_list
]
# Ordered first-wins for duplicate tool names across servers; keep tools.py tools.
self.load_registry_servers(
@@ -757,9 +751,7 @@ class ToolRegistry:
if preserve_existing_tools and mcp_tool.name in self._tools:
if log_collisions:
origin_server = (
self._find_mcp_origin_server_for_tool(mcp_tool.name) or "<existing>"
)
origin_server = self._find_mcp_origin_server_for_tool(mcp_tool.name) or "<existing>"
logger.warning(
"MCP tool '%s' from '%s' shadowed by '%s' (loaded first)",
mcp_tool.name,
@@ -788,17 +780,11 @@ class ToolRegistry:
base_context.update(exec_ctx)
# Only inject context params the tool accepts
filtered_context = {
k: v for k, v in base_context.items() if k in tool_params
}
filtered_context = {k: v for k, v in base_context.items() if k in tool_params}
# Strip context params from LLM inputs — the framework
# values are authoritative (prevents the LLM from passing
# e.g. data_dir="/data" and overriding the real path).
clean_inputs = {
k: v
for k, v in inputs.items()
if k not in registry_ref.CONTEXT_PARAMS
}
clean_inputs = {k: v for k, v in inputs.items() if k not in registry_ref.CONTEXT_PARAMS}
merged_inputs = {**clean_inputs, **filtered_context}
result = client_ref.call_tool(tool_name, merged_inputs)
# MCP client already extracts content (returns str
@@ -885,9 +871,7 @@ class ToolRegistry:
contents are already logged by `register_mcp_server`; this is just the
rollup so the resync path also gets a single anchor line.
"""
per_server_counts = {
server: len(names) for server, names in self._mcp_server_tools.items()
}
per_server_counts = {server: len(names) for server, names in self._mcp_server_tools.items()}
non_mcp_count = len(self._tools) - len(self._mcp_tool_names)
logger.info(
"ToolRegistry snapshot (%s): total=%d, mcp=%d, non_mcp=%d, per_server=%s",
@@ -958,11 +942,7 @@ class ToolRegistry:
adapter = CredentialStoreAdapter.default()
tool_provider_map = adapter.get_tool_provider_map()
live_providers = {
a.get("provider", "")
for a in adapter.get_all_account_info()
if a.get("provider")
}
live_providers = {a.get("provider", "") for a in adapter.get_all_account_info() if a.get("provider")}
except Exception:
logger.debug("Credential snapshot unavailable for MCP gate", exc_info=True)
core/framework/orchestrator/checkpoint_config.py
+1 -5
View File
@@ -50,11 +50,7 @@ class CheckpointConfig:
Returns:
True if should check for old checkpoints and prune them
"""
return (
self.enabled
and self.prune_every_n_nodes > 0
and nodes_executed % self.prune_every_n_nodes == 0
)
return self.enabled and self.prune_every_n_nodes > 0 and nodes_executed % self.prune_every_n_nodes == 0
# Default configuration for most agents
core/framework/orchestrator/context.py
+1 -3
View File
@@ -175,9 +175,7 @@ def _resolve_available_tools(
return always_tools
declared = set(node_spec.tools)
declared_tools = [
t for t in tools if t.name in declared and t.name not in _ALWAYS_AVAILABLE_TOOLS
]
declared_tools = [t for t in tools if t.name in declared and t.name not in _ALWAYS_AVAILABLE_TOOLS]
return always_tools + declared_tools
core/framework/orchestrator/context_handoff.py
+1 -5
View File
@@ -169,11 +169,7 @@ class ContextHandoff:
key_hint = ""
if output_keys:
key_hint = (
"\nThe following output keys are especially important: "
+ ", ".join(output_keys)
+ ".\n"
)
key_hint = "\nThe following output keys are especially important: " + ", ".join(output_keys) + ".\n"
system_prompt = (
"You are a concise summarizer. Given the conversation below, "
core/framework/orchestrator/edge.py
+5 -14
View File
@@ -186,8 +186,7 @@ class EdgeSpec(BaseModel):
expr_vars = {
k: repr(context[k])
for k in context
if k not in ("output", "buffer", "result", "true", "false")
and k in self.condition_expr
if k not in ("output", "buffer", "result", "true", "false") and k in self.condition_expr
}
logger.info(
" Edge %s: condition '%s' → %s (vars: %s)",
@@ -333,12 +332,8 @@ class GraphSpec(BaseModel):
default_factory=dict,
description="Named entry points for resuming execution. Format: {name: node_id}",
)
terminal_nodes: list[str] = Field(
default_factory=list, description="IDs of nodes that end execution"
)
pause_nodes: list[str] = Field(
default_factory=list, description="IDs of nodes that pause execution for HITL input"
)
terminal_nodes: list[str] = Field(default_factory=list, description="IDs of nodes that end execution")
pause_nodes: list[str] = Field(default_factory=list, description="IDs of nodes that pause execution for HITL input")
# Components
nodes: list[Any] = Field( # NodeSpec, but avoiding circular import
@@ -347,9 +342,7 @@ class GraphSpec(BaseModel):
edges: list[EdgeSpec] = Field(default_factory=list, description="All edge specifications")
# Data buffer keys
buffer_keys: list[str] = Field(
default_factory=list, description="Keys available in data buffer"
)
buffer_keys: list[str] = Field(default_factory=list, description="Keys available in data buffer")
# Default LLM settings
default_model: str = "claude-haiku-4-5-20251001"
@@ -557,9 +550,7 @@ class GraphSpec(BaseModel):
fan_outs = self.detect_fan_out_nodes()
for source_id, targets in fan_outs.items():
event_loop_targets = [
t
for t in targets
if self.get_node(t) and getattr(self.get_node(t), "node_type", "") == "event_loop"
t for t in targets if self.get_node(t) and getattr(self.get_node(t), "node_type", "") == "event_loop"
]
if len(event_loop_targets) > 1:
seen_keys: dict[str, str] = {}
core/framework/orchestrator/goal.py
+6 -18
View File
@@ -41,13 +41,9 @@ class SuccessCriterion(BaseModel):
id: str
description: str = Field(description="Human-readable description of what success looks like")
metric: str = Field(
description="How to measure: 'output_contains', 'output_equals', 'llm_judge', 'custom'"
)
metric: str = Field(description="How to measure: 'output_contains', 'output_equals', 'llm_judge', 'custom'")
# NEW: runtime evaluation type (separate from metric)
type: str = Field(
default="success_rate", description="Runtime evaluation type, e.g. 'success_rate'"
)
type: str = Field(default="success_rate", description="Runtime evaluation type, e.g. 'success_rate'")
target: Any = Field(description="The target value or condition")
weight: float = Field(default=1.0, ge=0.0, le=1.0, description="Relative importance (0-1)")
@@ -67,15 +63,9 @@ class Constraint(BaseModel):
id: str
description: str
constraint_type: str = Field(
description="Type: 'hard' (must not violate) or 'soft' (prefer not to violate)"
)
category: str = Field(
default="general", description="Category: 'time', 'cost', 'safety', 'scope', 'quality'"
)
check: str = Field(
default="", description="How to check: expression, function name, or 'llm_judge'"
)
constraint_type: str = Field(description="Type: 'hard' (must not violate) or 'soft' (prefer not to violate)")
category: str = Field(default="general", description="Category: 'time', 'cost', 'safety', 'scope', 'quality'")
check: str = Field(default="", description="How to check: expression, function name, or 'llm_judge'")
model_config = {"extra": "allow"}
@@ -142,9 +132,7 @@ class Goal(BaseModel):
# Input/output schema
input_schema: dict[str, Any] = Field(default_factory=dict, description="Expected input format")
output_schema: dict[str, Any] = Field(
default_factory=dict, description="Expected output format"
)
output_schema: dict[str, Any] = Field(default_factory=dict, description="Expected output format")
# Versioning for evolution
version: str = "1.0.0"
core/framework/orchestrator/node.py
+5 -13
View File
@@ -129,15 +129,13 @@ class NodeSpec(BaseModel):
input_schema: dict[str, dict] = Field(
default_factory=dict,
description=(
"Optional schema for input validation. "
"Format: {key: {type: 'string', required: True, description: '...'}}"
"Optional schema for input validation. Format: {key: {type: 'string', required: True, description: '...'}}"
),
)
output_schema: dict[str, dict] = Field(
default_factory=dict,
description=(
"Optional schema for output validation. "
"Format: {key: {type: 'dict', required: True, description: '...'}}"
"Optional schema for output validation. Format: {key: {type: 'dict', required: True, description: '...'}}"
),
)
@@ -153,19 +151,13 @@ class NodeSpec(BaseModel):
"'none' = no tools at all."
),
)
model: str | None = Field(
default=None, description="Specific model to use (defaults to graph default)"
)
model: str | None = Field(default=None, description="Specific model to use (defaults to graph default)")
# For function nodes
function: str | None = Field(
default=None, description="Function name or path for function nodes"
)
function: str | None = Field(default=None, description="Function name or path for function nodes")
# For router nodes
routes: dict[str, str] = Field(
default_factory=dict, description="Condition -> target_node_id mapping for routers"
)
routes: dict[str, str] = Field(default_factory=dict, description="Condition -> target_node_id mapping for routers")
# Retry behavior
max_retries: int = Field(default=3)
core/framework/orchestrator/node_worker.py
+7 -20
View File
@@ -379,9 +379,7 @@ class NodeWorker:
# Failure
if attempt + 1 < total_attempts:
gc.retry_counts[self.node_spec.id] = (
gc.retry_counts.get(self.node_spec.id, 0) + 1
)
gc.retry_counts[self.node_spec.id] = gc.retry_counts.get(self.node_spec.id, 0) + 1
gc.nodes_with_retries.add(self.node_spec.id)
delay = 1.0 * (2**attempt)
logger.warning(
@@ -411,9 +409,7 @@ class NodeWorker:
except Exception as exc:
if attempt + 1 < total_attempts:
gc.retry_counts[self.node_spec.id] = (
gc.retry_counts.get(self.node_spec.id, 0) + 1
)
gc.retry_counts[self.node_spec.id] = gc.retry_counts.get(self.node_spec.id, 0) + 1
gc.nodes_with_retries.add(self.node_spec.id)
delay = 1.0 * (2**attempt)
logger.warning(
@@ -469,9 +465,7 @@ class NodeWorker:
if len(conditionals) > 1:
max_prio = max(e.priority for e in conditionals)
traversable = [
e
for e in traversable
if e.condition != EdgeCondition.CONDITIONAL or e.priority == max_prio
e for e in traversable if e.condition != EdgeCondition.CONDITIONAL or e.priority == max_prio
]
# When parallel execution is disabled, follow first match only (sequential)
@@ -541,9 +535,7 @@ class NodeWorker:
logger.warning("Worker %s output validation warnings: %s", node_spec.id, errors)
# Determine if this worker is a fan-out branch
is_fanout_branch = any(
tag.via_branch == node_spec.id for tag in self._inherited_fan_out_tags
)
is_fanout_branch = any(tag.via_branch == node_spec.id for tag in self._inherited_fan_out_tags)
# Collect keys to write: declared output_keys + any extra output items
# (for fan-out branches, all output items need conflict checking)
@@ -642,9 +634,7 @@ class NodeWorker:
self._node_impl = node
return node
raise RuntimeError(
f"No implementation for node '{self.node_spec.id}' (type: {self.node_spec.node_type})"
)
raise RuntimeError(f"No implementation for node '{self.node_spec.id}' (type: {self.node_spec.node_type})")
def _build_node_context(self) -> NodeContext:
"""Build NodeContext for this worker's execution."""
@@ -749,9 +739,7 @@ class NodeWorker:
inherited_conversation=gc.continuous_conversation,
narrative=narrative,
)
gc.continuous_conversation.update_system_prompt(
build_system_prompt_for_node_context(next_ctx)
)
gc.continuous_conversation.update_system_prompt(build_system_prompt_for_node_context(next_ctx))
gc.continuous_conversation.set_current_phase(next_spec.id)
buffer_items, data_files = self._prepare_transition_payload()
@@ -799,8 +787,7 @@ class NodeWorker:
file_path.write_text(write_content, encoding="utf-8")
file_size = file_path.stat().st_size
buffer_items[key] = (
f"[Saved to '{filename}' ({file_size:,} bytes). "
f"Use read_file(path='{filename}') to access.]"
f"[Saved to '{filename}' ({file_size:,} bytes). Use read_file(path='{filename}') to access.]"
)
continue
except Exception:
core/framework/orchestrator/orchestrator.py
+35 -110
View File
@@ -202,9 +202,7 @@ class Orchestrator:
self.validator = OutputValidator()
self.logger = logging.getLogger(__name__)
self.logger.debug(
"[Orchestrator.__init__] Created with"
" stream_id=%s, execution_id=%s,"
" initial node_registry keys: %s",
"[Orchestrator.__init__] Created with stream_id=%s, execution_id=%s, initial node_registry keys: %s",
stream_id,
execution_id,
list(self.node_registry.keys()),
@@ -347,8 +345,7 @@ class Orchestrator:
missing = [t for t in declared if t not in available_tool_names]
if missing:
self.logger.warning(
"Node '%s' (id=%s) declares %d tools not in this runtime; "
"stripping them and continuing: %s",
"Node '%s' (id=%s) declares %d tools not in this runtime; stripping them and continuing: %s",
node.name,
node.id,
len(missing),
@@ -391,10 +388,7 @@ class Orchestrator:
lines.append(f"[tool result]: {c}")
elif m.role == "assistant" and m.tool_calls:
names = [tc.get("function", {}).get("name", "?") for tc in m.tool_calls]
lines.append(
f"[assistant (calls: {', '.join(names)})]: "
f"{m.content[:200] if m.content else ''}"
)
lines.append(f"[assistant (calls: {', '.join(names)})]: {m.content[:200] if m.content else ''}")
else:
lines.append(f"[{m.role}]: {m.content}")
formatted = "\n\n".join(lines)
@@ -565,8 +559,7 @@ class Orchestrator:
# [RESTORED] Type safety check
if not isinstance(buffer_data, dict):
self.logger.warning(
f"⚠️ Invalid data buffer type in session state: "
f"{type(buffer_data).__name__}, expected dict"
f"⚠️ Invalid data buffer type in session state: {type(buffer_data).__name__}, expected dict"
)
else:
# Restore buffer from previous session.
@@ -590,8 +583,7 @@ class Orchestrator:
# contains all state including the original input, and re-writing
# input_data would overwrite intermediate results with stale values.
_is_resuming = bool(
session_state
and (session_state.get("paused_at") or session_state.get("resume_from_checkpoint"))
session_state and (session_state.get("paused_at") or session_state.get("resume_from_checkpoint"))
)
if input_data and not _is_resuming:
for key, value in input_data.items():
@@ -616,11 +608,7 @@ class Orchestrator:
# If resuming at a specific node (paused_at), that node was counted
# but never completed, so decrement its count
paused_at = session_state.get("paused_at")
if (
paused_at
and paused_at in node_visit_counts
and node_visit_counts[paused_at] > 0
):
if paused_at and paused_at in node_visit_counts and node_visit_counts[paused_at] > 0:
old_count = node_visit_counts[paused_at]
node_visit_counts[paused_at] -= 1
self.logger.info(
@@ -636,10 +624,7 @@ class Orchestrator:
checkpoint = await checkpoint_store.load_checkpoint(checkpoint_id)
if checkpoint:
self.logger.info(
f"🔄 Resuming from checkpoint: {checkpoint_id} "
f"(node: {checkpoint.current_node})"
)
self.logger.info(f"🔄 Resuming from checkpoint: {checkpoint_id} (node: {checkpoint.current_node})")
checkpoint_run_id = checkpoint.run_id or LEGACY_RUN_ID
self._run_id = checkpoint_run_id
@@ -648,9 +633,7 @@ class Orchestrator:
buffer.write(key, value, validate=False)
# Start from checkpoint's next node or current node
current_node_id = (
checkpoint.next_node or checkpoint.current_node or graph.entry_node
)
current_node_id = checkpoint.next_node or checkpoint.current_node or graph.entry_node
# Restore execution path
path.extend(checkpoint.execution_path)
@@ -660,16 +643,11 @@ class Orchestrator:
f"resuming at node: {current_node_id}"
)
else:
self.logger.warning(
f"Checkpoint {checkpoint_id} not found, resuming from normal entry point"
)
self.logger.warning(f"Checkpoint {checkpoint_id} not found, resuming from normal entry point")
current_node_id = graph.get_entry_point(session_state)
except Exception as e:
self.logger.error(
f"Failed to load checkpoint {checkpoint_id}: {e}, "
f"resuming from normal entry point"
)
self.logger.error(f"Failed to load checkpoint {checkpoint_id}: {e}, resuming from normal entry point")
current_node_id = graph.get_entry_point(session_state)
else:
current_node_id = graph.get_entry_point(session_state)
@@ -757,20 +735,14 @@ class Orchestrator:
"human_input": "event_loop", # Use queen interaction / escalation instead
}
def _get_node_implementation(
self, node_spec: NodeSpec, cleanup_llm_model: str | None = None
) -> NodeProtocol:
def _get_node_implementation(self, node_spec: NodeSpec, cleanup_llm_model: str | None = None) -> NodeProtocol:
"""Get or create a node implementation."""
# Check registry first
if node_spec.id in self.node_registry:
logger.debug(
"[Orchestrator._get_node_implementation] Found node '%s' in registry", node_spec.id
)
logger.debug("[Orchestrator._get_node_implementation] Found node '%s' in registry", node_spec.id)
return self.node_registry[node_spec.id]
logger.debug(
"[Orchestrator._get_node_implementation]"
" Node '%s' not in registry (keys: %s),"
" creating new",
"[Orchestrator._get_node_implementation] Node '%s' not in registry (keys: %s), creating new",
node_spec.id,
list(self.node_registry.keys()),
)
@@ -840,9 +812,7 @@ class Orchestrator:
# Cache so inject_event() is reachable for queen interaction and escalation routing
self.node_registry[node_spec.id] = node
logger.debug(
"[Orchestrator._get_node_implementation]"
" Cached node '%s' in node_registry,"
" registry now has keys: %s",
"[Orchestrator._get_node_implementation] Cached node '%s' in node_registry, registry now has keys: %s",
node_spec.id,
list(self.node_registry.keys()),
)
@@ -925,9 +895,7 @@ class Orchestrator:
if len(conditionals) > 1:
max_prio = max(e.priority for e in conditionals)
traversable = [
e
for e in traversable
if e.condition != EdgeCondition.CONDITIONAL or e.priority == max_prio
e for e in traversable if e.condition != EdgeCondition.CONDITIONAL or e.priority == max_prio
]
return traversable
@@ -1090,9 +1058,7 @@ class Orchestrator:
execution_id=self._execution_id,
)
self.logger.info(
f" ▶ Branch {node_spec.name}: executing (attempt {attempt + 1})"
)
self.logger.info(f" ▶ Branch {node_spec.name}: executing (attempt {attempt + 1})")
result = await node_impl.execute(ctx)
last_result = result
@@ -1153,19 +1119,13 @@ class Orchestrator:
)
return branch, result
self.logger.warning(
f" ↻ Branch {node_spec.name}: "
f"retry {attempt + 1}/{effective_max_retries}"
)
self.logger.warning(f" ↻ Branch {node_spec.name}: retry {attempt + 1}/{effective_max_retries}")
# All retries exhausted
branch.status = "failed"
branch.error = last_result.error if last_result else "Unknown error"
branch.result = last_result
self.logger.error(
f" ✗ Branch {node_spec.name}: "
f"failed after {effective_max_retries} attempts"
)
self.logger.error(f" ✗ Branch {node_spec.name}: failed after {effective_max_retries} attempts")
return branch, last_result
except Exception as e:
@@ -1208,10 +1168,7 @@ class Orchestrator:
# Branch timed out
branch.status = "timed_out"
branch.error = f"Branch timed out after {timeout}s"
self.logger.warning(
f" ⏱ Branch {graph.get_node(branch.node_id).name}: "
f"timed out after {timeout}s"
)
self.logger.warning(f" ⏱ Branch {graph.get_node(branch.node_id).name}: timed out after {timeout}s")
path.append(branch.node_id)
failed_branches.append(branch)
elif isinstance(result, Exception):
@@ -1235,13 +1192,9 @@ class Orchestrator:
if self._parallel_config.on_branch_failure == "fail_all":
raise RuntimeError(f"Parallel execution failed: branches {failed_names} failed")
elif self._parallel_config.on_branch_failure == "continue_others":
self.logger.warning(
f"⚠ Some branches failed ({failed_names}), continuing with successful ones"
)
self.logger.warning(f"⚠ Some branches failed ({failed_names}), continuing with successful ones")
self.logger.info(
f" ⑃ Fan-out complete: {len(branch_results)}/{len(branches)} branches succeeded"
)
self.logger.info(f" ⑃ Fan-out complete: {len(branch_results)}/{len(branches)} branches succeeded")
return branch_results, total_tokens, total_latency
def register_node(self, node_id: str, implementation: NodeProtocol) -> None:
@@ -1432,15 +1385,10 @@ class Orchestrator:
return True
if not terminal_worker_ids:
# No terminals: check if all workers are done
return all(
w.lifecycle in (WorkerLifecycle.COMPLETED, WorkerLifecycle.FAILED)
for w in workers.values()
)
return all(w.lifecycle in (WorkerLifecycle.COMPLETED, WorkerLifecycle.FAILED) for w in workers.values())
if any(w.lifecycle == WorkerLifecycle.RUNNING for w in workers.values()):
return False
return any(
tid in completed_terminals or tid in failed_workers for tid in terminal_worker_ids
)
return any(tid in completed_terminals or tid in failed_workers for tid in terminal_worker_ids)
def _mark_quiescent_terminal_failure() -> bool:
nonlocal execution_error
@@ -1448,22 +1396,15 @@ class Orchestrator:
return False
if any(w.lifecycle == WorkerLifecycle.RUNNING for w in workers.values()):
return False
if any(
tid in completed_terminals or tid in failed_workers for tid in terminal_worker_ids
):
if any(tid in completed_terminals or tid in failed_workers for tid in terminal_worker_ids):
return False
execution_error = (
"Worker execution ended before terminal nodes completed: "
f"{sorted(terminal_worker_ids)}"
)
execution_error = f"Worker execution ended before terminal nodes completed: {sorted(terminal_worker_ids)}"
self.logger.error(execution_error)
return True
# Track fan-out branch workers for per-branch timeout enforcement
_fanout_branch_tasks: dict[str, asyncio.Task] = {} # worker_id → timeout-wrapper task
branch_timeout = (
self._parallel_config.branch_timeout_seconds if self._parallel_config else 300.0
)
branch_timeout = self._parallel_config.branch_timeout_seconds if self._parallel_config else 300.0
def _route_activation(
activation: Activation,
@@ -1498,9 +1439,7 @@ class Orchestrator:
target_worker.activate(inherited_tags=activation.fan_out_tags)
if target_worker._task is not None:
# Fan-out branch: wrap with timeout
is_fanout_branch = any(
tag.via_branch == activation.target_id for tag in activation.fan_out_tags
)
is_fanout_branch = any(tag.via_branch == activation.target_id for tag in activation.fan_out_tags)
if is_fanout_branch and branch_timeout > 0:
timed_task = asyncio.ensure_future(
asyncio.wait_for(target_worker._task, timeout=branch_timeout)
@@ -1555,9 +1494,7 @@ class Orchestrator:
if completion.conversation is not None:
gc.continuous_conversation = completion.conversation
self.logger.info(
f" ✓ Worker completed: {worker_id} ({len(activations)} outgoing activation(s))"
)
self.logger.info(f" ✓ Worker completed: {worker_id} ({len(activations)} outgoing activation(s))")
# Route activations to target workers
for activation in activations:
@@ -1598,9 +1535,7 @@ class Orchestrator:
completion_event.set()
# Subscribe to events (only if event bus has subscribe capability)
has_event_subscription = self._event_bus is not None and hasattr(
self._event_bus, "subscribe"
)
has_event_subscription = self._event_bus is not None and hasattr(self._event_bus, "subscribe")
if has_event_subscription:
sub_completed = self._event_bus.subscribe(
event_types=[EventType.WORKER_COMPLETED],
@@ -1642,14 +1577,12 @@ class Orchestrator:
)
if unresolved_terminals:
execution_error = (
"Worker execution ended before terminal nodes completed: "
f"{unresolved_terminals}"
f"Worker execution ended before terminal nodes completed: {unresolved_terminals}"
)
self.logger.error(execution_error)
else:
execution_error = (
"Worker execution ended before all workers reached "
"a terminal lifecycle state"
"Worker execution ended before all workers reached a terminal lifecycle state"
)
self.logger.error(execution_error)
break
@@ -1680,10 +1613,7 @@ class Orchestrator:
task_error = exc
# Check for fan-out branch timeout
if (
isinstance(task_error, asyncio.TimeoutError)
and wid in _fanout_branch_tasks
):
if isinstance(task_error, asyncio.TimeoutError) and wid in _fanout_branch_tasks:
error = f"Branch failed (timed out after {branch_timeout}s)"
failed_workers[wid] = error
worker.lifecycle = WorkerLifecycle.FAILED
@@ -1727,10 +1657,7 @@ class Orchestrator:
src_spec = graph.get_node(wid)
if src_spec and src_spec.tools:
for t in self.tools:
if (
t.name in src_spec.tools
and t.name not in gc.cumulative_tool_names
):
if t.name in src_spec.tools and t.name not in gc.cumulative_tool_names:
gc.cumulative_tools.append(t)
gc.cumulative_tool_names.add(t.name)
if src_spec and src_spec.output_keys:
@@ -1741,8 +1668,7 @@ class Orchestrator:
gc.continuous_conversation = completion_conversation
self.logger.info(
f" ✓ Worker completed: {wid} "
f"({len(outgoing_activations)} outgoing activation(s))"
f" ✓ Worker completed: {wid} ({len(outgoing_activations)} outgoing activation(s))"
)
# Route activations
@@ -1787,8 +1713,7 @@ class Orchestrator:
error = str(task_error)
else:
error = (
"Worker task completed without publishing a completion "
f"(lifecycle={worker.lifecycle})"
f"Worker task completed without publishing a completion (lifecycle={worker.lifecycle})"
)
failed_workers[wid] = error
core/framework/orchestrator/prompt_composer.py
+2 -5
View File
@@ -97,15 +97,12 @@ def build_transition_marker(
file_path = data_path / filename
try:
write_content = (
json.dumps(value, indent=2, ensure_ascii=False)
if isinstance(value, (dict, list))
else str(value)
json.dumps(value, indent=2, ensure_ascii=False) if isinstance(value, (dict, list)) else str(value)
)
file_path.write_text(write_content, encoding="utf-8")
file_size = file_path.stat().st_size
buffer_items[key] = (
f"[Saved to '{filename}' ({file_size:,} bytes). "
f"Use read_file(path='{filename}') to access.]"
f"[Saved to '{filename}' ({file_size:,} bytes). Use read_file(path='{filename}') to access.]"
)
except Exception:
buffer_items[key] = val_str[:300] + "..."
core/framework/orchestrator/prompting.py
+4 -11
View File
@@ -177,18 +177,12 @@ def build_prompt_spec_from_node_context(
# Tool-gated pre-activation: inject full body of default skills whose
# trigger tools are present in this node's tool list (e.g. browser_*
# pulls in hive.browser-automation).
tool_names = [
getattr(t, "name", "") for t in (getattr(ctx, "available_tools", None) or [])
]
skills_catalog_prompt = augment_catalog_for_tools(
ctx.skills_catalog_prompt or "", tool_names
)
tool_names = [getattr(t, "name", "") for t in (getattr(ctx, "available_tools", None) or [])]
skills_catalog_prompt = augment_catalog_for_tools(ctx.skills_catalog_prompt or "", tool_names)
return NodePromptSpec(
identity_prompt=ctx.identity_prompt or "",
focus_prompt=focus_prompt
if focus_prompt is not None
else (ctx.node_spec.system_prompt or ""),
focus_prompt=focus_prompt if focus_prompt is not None else (ctx.node_spec.system_prompt or ""),
narrative=narrative if narrative is not None else (ctx.narrative or ""),
accounts_prompt=ctx.accounts_prompt or "",
skills_catalog_prompt=skills_catalog_prompt,
@@ -299,8 +293,7 @@ def build_transition_message(spec: TransitionSpec) -> str:
if spec.data_files:
sections.append(
"\nData files (use read_file to access):\n"
+ "\n".join(f" {entry}" for entry in spec.data_files)
"\nData files (use read_file to access):\n" + "\n".join(f" {entry}" for entry in spec.data_files)
)
if spec.cumulative_tool_names:
core/framework/orchestrator/safe_eval.py
+1 -5
View File
@@ -169,11 +169,7 @@ class SafeEvalVisitor(ast.NodeVisitor):
return tuple(self.visit(elt) for elt in node.elts)
def visit_Dict(self, node: ast.Dict) -> dict:
return {
self.visit(k): self.visit(v)
for k, v in zip(node.keys, node.values, strict=False)
if k is not None
}
return {self.visit(k): self.visit(v) for k, v in zip(node.keys, node.values, strict=False) if k is not None}
# --- Operations ---
def visit_BinOp(self, node: ast.BinOp) -> Any:
core/framework/orchestrator/validator.py
+2 -6
View File
@@ -120,9 +120,7 @@ class OutputValidator:
nullable_keys = nullable_keys or []
if not isinstance(output, dict):
return ValidationResult(
success=False, errors=[f"Output is not a dict, got {type(output).__name__}"]
)
return ValidationResult(success=False, errors=[f"Output is not a dict, got {type(output).__name__}"])
for key in expected_keys:
if key not in output:
@@ -237,9 +235,7 @@ class OutputValidator:
# Check for overly long values
if len(value) > max_length:
errors.append(
f"Output key '{key}' exceeds max length ({len(value)} > {max_length})"
)
errors.append(f"Output key '{key}' exceeds max length ({len(value)} > {max_length})")
return ValidationResult(success=len(errors) == 0, errors=errors)
core/framework/pipeline/stages/cost_guard.py
+1 -3
View File
@@ -27,8 +27,6 @@ class CostGuardStage(PipelineStage):
if estimated > self._budget:
return PipelineResult(
action="reject",
rejection_reason=(
f"Estimated cost ${estimated:.4f} exceeds budget ${self._budget:.4f}"
),
rejection_reason=(f"Estimated cost ${estimated:.4f} exceeds budget ${self._budget:.4f}"),
)
return PipelineResult(action="continue")
core/framework/pipeline/stages/input_validation.py
+1 -2
View File
@@ -40,8 +40,7 @@ class InputValidationStage(PipelineStage):
return PipelineResult(
action="reject",
rejection_reason=(
f"Input key '{key}' has type {type(value).__name__}, "
f"expected {expected_type.__name__}"
f"Input key '{key}' has type {type(value).__name__}, expected {expected_type.__name__}"
),
)
return PipelineResult(action="continue")
core/framework/pipeline/stages/rate_limit.py
+1 -3
View File
@@ -35,9 +35,7 @@ class RateLimitStage(PipelineStage):
if len(self._timestamps[key]) >= self._max_rpm:
return PipelineResult(
action="reject",
rejection_reason=(
f"Rate limit exceeded: {self._max_rpm} req/min for session '{session_id}'"
),
rejection_reason=(f"Rate limit exceeded: {self._max_rpm} req/min for session '{session_id}'"),
)
self._timestamps[key].append(now)
return PipelineResult(action="continue")
core/framework/schemas/goal.py
+4 -12
View File
@@ -25,9 +25,7 @@ class GoalStatus(StrEnum):
class SuccessCriterion(BaseModel):
id: str
description: str = Field(description="Human-readable description of what success looks like")
metric: str = Field(
description="How to measure: 'output_contains', 'output_equals', 'llm_judge', 'custom'"
)
metric: str = Field(description="How to measure: 'output_contains', 'output_equals', 'llm_judge', 'custom'")
type: str = Field(default="success_rate", description="Runtime evaluation type")
target: Any = Field(description="The target value or condition")
weight: float = Field(default=1.0, ge=0.0, le=1.0, description="Relative importance (0-1)")
@@ -39,15 +37,9 @@ class SuccessCriterion(BaseModel):
class Constraint(BaseModel):
id: str
description: str
constraint_type: str = Field(
description="Type: 'hard' (must not violate) or 'soft' (prefer not to violate)"
)
category: str = Field(
default="general", description="Category: 'time', 'cost', 'safety', 'scope', 'quality'"
)
check: str = Field(
default="", description="How to check: expression, function name, or 'llm_judge'"
)
constraint_type: str = Field(description="Type: 'hard' (must not violate) or 'soft' (prefer not to violate)")
category: str = Field(default="general", description="Category: 'time', 'cost', 'safety', 'scope', 'quality'")
check: str = Field(default="", description="How to check: expression, function name, or 'llm_judge'")
model_config = {"extra": "allow"}
core/framework/schemas/session_state.py
+2 -6
View File
@@ -237,9 +237,7 @@ class SessionState(BaseModel):
progress=SessionProgress(
current_node=result.paused_at or (result.path[-1] if result.path else None),
paused_at=result.paused_at,
resume_from=result.session_state.get("resume_from")
if result.session_state
else None,
resume_from=result.session_state.get("resume_from") if result.session_state else None,
steps_executed=result.steps_executed,
total_tokens=result.total_tokens,
total_latency_ms=result.total_latency_ms,
@@ -256,9 +254,7 @@ class SessionState(BaseModel):
error=result.error,
output=result.output,
),
data_buffer=result.session_state.get(
"data_buffer", result.session_state.get("memory", {})
)
data_buffer=result.session_state.get("data_buffer", result.session_state.get("memory", {}))
if result.session_state
else {},
input_data=input_data or {},
core/framework/server/app.py
+5 -12
View File
@@ -56,8 +56,7 @@ def validate_agent_path(agent_path: str | Path) -> Path:
if resolved.is_relative_to(root) and resolved != root:
return resolved
raise ValueError(
"agent_path must be inside an allowed directory "
"(~/.hive/colonies/, exports/, examples/, or ~/.hive/agents/)"
"agent_path must be inside an allowed directory (~/.hive/colonies/, exports/, examples/, or ~/.hive/agents/)"
)
@@ -186,9 +185,7 @@ async def handle_browser_status(request: web.Request) -> web.Response:
status_port = bridge_port + 1
try:
reader, writer = await asyncio.wait_for(
asyncio.open_connection("127.0.0.1", status_port), timeout=0.5
)
reader, writer = await asyncio.wait_for(asyncio.open_connection("127.0.0.1", status_port), timeout=0.5)
writer.write(b"GET /status HTTP/1.0\r\nHost: 127.0.0.1\r\n\r\n")
await writer.drain()
raw = await asyncio.wait_for(reader.read(512), timeout=0.5)
@@ -232,9 +229,7 @@ def create_app(model: str | None = None) -> web.Application:
from framework.credentials.key_storage import generate_and_save_credential_key
generate_and_save_credential_key()
logger.info(
"Generated and persisted HIVE_CREDENTIAL_KEY to ~/.hive/secrets/credential_key"
)
logger.info("Generated and persisted HIVE_CREDENTIAL_KEY to ~/.hive/secrets/credential_key")
except Exception as exc:
logger.warning("Could not auto-persist HIVE_CREDENTIAL_KEY: %s", exc)
@@ -274,9 +269,7 @@ def create_app(model: str | None = None) -> web.Application:
log_collisions=True,
max_tools=selection_max_tools,
)
logger.info(
"Pre-loaded queen tool registry with %d tools", len(_queen_tool_registry.get_tools())
)
logger.info("Pre-loaded queen tool registry with %d tools", len(_queen_tool_registry.get_tools()))
except Exception as e:
logger.warning("Failed to pre-load queen tool registry: %s", e)
@@ -297,11 +290,11 @@ def create_app(model: str | None = None) -> web.Application:
from framework.server.routes_credentials import register_routes as register_credential_routes
from framework.server.routes_events import register_routes as register_event_routes
from framework.server.routes_execution import register_routes as register_execution_routes
from framework.server.routes_workers import register_routes as register_worker_routes
from framework.server.routes_logs import register_routes as register_log_routes
from framework.server.routes_messages import register_routes as register_message_routes
from framework.server.routes_queens import register_routes as register_queen_routes
from framework.server.routes_sessions import register_routes as register_session_routes
from framework.server.routes_workers import register_routes as register_worker_routes
register_config_routes(app)
register_credential_routes(app)
core/framework/server/queen_orchestrator.py
+28 -47
View File
@@ -13,6 +13,8 @@ from pathlib import Path
from typing import TYPE_CHECKING, Any
if TYPE_CHECKING:
from framework.agent_loop.internals.types import HookContext, HookResult
from framework.loader.tool_registry import ToolRegistry
from framework.server.session_manager import Session
logger = logging.getLogger(__name__)
@@ -49,7 +51,7 @@ def install_worker_escalation_routing(
# Defensive: ignore any stray non-worker origin (e.g. queen).
if not stream_id.startswith("worker:"):
return
worker_id = stream_id[len("worker:"):]
worker_id = stream_id[len("worker:") :]
data = event.data or {}
request_id = data.get("request_id")
reason = str(data.get("reason", "")).strip()
@@ -64,8 +66,7 @@ def install_worker_escalation_routing(
try:
await runtime.inject_input(
worker_id,
"[QUEEN_REPLY] queue_full — queen inbox saturated; "
"proceed with best judgment or retry later.",
"[QUEEN_REPLY] queue_full — queen inbox saturated; proceed with best judgment or retry later.",
)
except Exception:
logger.warning(
@@ -100,24 +101,16 @@ def install_worker_escalation_routing(
lines.append(context_text)
if request_id:
lines.append(
"Use reply_to_worker(request_id, reply) to unblock, "
"or list_worker_questions() to see all pending."
"Use reply_to_worker(request_id, reply) to unblock, or list_worker_questions() to see all pending."
)
else:
lines.append(
"No request_id — use inject_message(content=...) to relay "
"guidance manually."
)
lines.append("No request_id — use inject_message(content=...) to relay guidance manually.")
handoff = "\n".join(lines)
# Fallback: if the queen loop has gone away, publish a
# CLIENT_INPUT_REQUESTED so the human sees the question and the
# worker does not wedge.
queen_node = (
session.queen_executor.node_registry.get("queen")
if session.queen_executor is not None
else None
)
queen_node = session.queen_executor.node_registry.get("queen") if session.queen_executor is not None else None
if queen_node is None or not hasattr(queen_node, "inject_event"):
if session.event_bus is not None:
await session.event_bus.emit_client_input_requested(
@@ -141,9 +134,7 @@ def install_worker_escalation_routing(
filter_colony=runtime.colony_id,
)
except Exception:
logger.warning(
"Failed to install colony-scoped escalation sub", exc_info=True
)
logger.warning("Failed to install colony-scoped escalation sub", exc_info=True)
# fall through to session bus
if session.event_bus is None:
return None
@@ -174,14 +165,12 @@ def _build_credentials_provider() -> Any:
def _provider() -> str:
now = time.monotonic()
if (
state["cached"]
and (now - state["cached_at"]) < _CREDENTIALS_BLOCK_TTL_SECONDS
):
if state["cached"] and (now - state["cached_at"]) < _CREDENTIALS_BLOCK_TTL_SECONDS:
return state["cached"]
try:
from aden_tools.credentials.store_adapter import CredentialStoreAdapter
from framework.orchestrator.prompting import build_accounts_prompt
adapter = CredentialStoreAdapter.default()
@@ -313,8 +302,8 @@ async def create_queen(
_shared_building_knowledge,
finalize_queen_prompt,
)
from framework.llm.capabilities import supports_image_tool_results
from framework.host.event_bus import AgentEvent, EventType
from framework.llm.capabilities import supports_image_tool_results
from framework.loader.mcp_registry import MCPRegistry
from framework.loader.tool_registry import ToolRegistry
from framework.tools.queen_lifecycle_tools import (
@@ -326,9 +315,7 @@ async def create_queen(
# Use pre-loaded cached registry if available (fast path)
if tool_registry is not None:
queen_registry = tool_registry
logger.info(
"Queen: using pre-loaded tool registry with %d tools", len(queen_registry.get_tools())
)
logger.info("Queen: using pre-loaded tool registry with %d tools", len(queen_registry.get_tools()))
else:
# Build fresh (slow path - for backwards compatibility)
queen_registry = ToolRegistry()
@@ -456,13 +443,9 @@ async def create_queen(
# Independent phase gets core tools + all MCP tools not claimed by any
# other phase (coder-tools file I/O, gcu-tools browser, etc.).
all_phase_names = (
planning_names | building_names | staging_names | running_names | editing_names
)
all_phase_names = planning_names | building_names | staging_names | running_names | editing_names
mcp_tools = [t for t in queen_tools if t.name not in all_phase_names]
phase_state.independent_tools = [
t for t in queen_tools if t.name in independent_names
] + mcp_tools
phase_state.independent_tools = [t for t in queen_tools if t.name in independent_names] + mcp_tools
logger.info(
"Queen: independent tools: %s",
sorted(t.name for t in phase_state.independent_tools),
@@ -494,9 +477,7 @@ async def create_queen(
# Resolve vision-only prompt sections based on the session's LLM.
# session.llm is immutable for the session's lifetime, so this check
# is stable — prompts never need to be recomposed mid-session.
_has_vision = bool(
session.llm and supports_image_tool_results(getattr(session.llm, "model", ""))
)
_has_vision = bool(session.llm and supports_image_tool_results(getattr(session.llm, "model", "")))
_planning_body = (
_queen_character_core
@@ -625,6 +606,14 @@ async def create_queen(
)
async def _queen_identity_hook(ctx: HookContext) -> HookResult | None:
from framework.agent_loop.internals.types import HookResult
from framework.agents.queen.queen_profiles import (
ensure_default_queens,
format_queen_identity_prompt,
load_queen_profile,
select_queen,
)
ensure_default_queens()
trigger = ctx.trigger or ""
# If the session was pre-bound to a queen (user clicked a specific
@@ -676,18 +665,12 @@ async def create_queen(
try:
_meta = _json.loads(_meta_path.read_text(encoding="utf-8"))
_meta["queen_id"] = queen_id
_meta_path.write_text(
_json.dumps(_meta, ensure_ascii=False), encoding="utf-8"
)
_meta_path.write_text(_json.dumps(_meta, ensure_ascii=False), encoding="utf-8")
except (OSError, _json.JSONDecodeError):
pass
# Re-point event bus log to new location, preserving offset
_offset = getattr(
session.event_bus, "_session_log_iteration_offset", 0
)
session.event_bus.set_session_log(
_new_dir / "events.jsonl", iteration_offset=_offset
)
_offset = getattr(session.event_bus, "_session_log_iteration_offset", 0)
session.event_bus.set_session_log(_new_dir / "events.jsonl", iteration_offset=_offset)
if _session_event_bus is not None:
await _session_event_bus.publish(
@@ -742,7 +725,7 @@ async def create_queen(
logger.debug("Queen: tools not yet available (registered on worker load): %s", missing)
node_updates["tools"] = available_tools
adjusted_node = _orig_node.model_copy(update=node_updates)
_orig_node.model_copy(update=node_updates)
# Determine session mode:
# - RESTORE: Resume cold session with history, no initial prompt -> wait for user
@@ -897,9 +880,7 @@ async def create_queen(
# bootstrap: if the frontend doesn't pass initial_prompt, we must
# NOT invent a phantom "Hello" — that used to concatenate with the
# real first chat message and confuse the model.
ctx.input_data = {
"user_request": None if _is_restore_mode else (initial_prompt or None)
}
ctx.input_data = {"user_request": None if _is_restore_mode else (initial_prompt or None)}
# Publish the initial prompt as a CLIENT_INPUT_RECEIVED event so
# it appears in the SSE stream and persists to events.jsonl for
core/framework/server/routes_config.py
+4 -12
View File
@@ -281,9 +281,7 @@ def _get_subscription_token(sub_id: str) -> str | None:
return None
def _hot_swap_sessions(
request: web.Request, full_model: str, api_key: str | None, api_base: str | None
) -> int:
def _hot_swap_sessions(request: web.Request, full_model: str, api_key: str | None, api_base: str | None) -> int:
"""Hot-swap the LLM on all running sessions. Returns count of swapped sessions.
Also refreshes the SessionManager's default model so that subsequent
@@ -363,9 +361,7 @@ async def handle_update_llm_config(request: web.Request) -> web.Response:
# ── Subscription mode ────────────────────────────────────────
sub = _SUBSCRIPTION_MAP.get(subscription_id)
if not sub:
return web.json_response(
{"error": f"Unknown subscription: {subscription_id}"}, status=400
)
return web.json_response({"error": f"Unknown subscription: {subscription_id}"}, status=400)
preset = get_preset(subscription_id)
# Subscriptions use the fixed model from their preset (no model switching)
@@ -432,9 +428,7 @@ async def handle_update_llm_config(request: web.Request) -> web.Response:
provider = body.get("provider")
model = body.get("model")
if not provider or not model:
return web.json_response(
{"error": "Both 'provider' and 'model' are required"}, status=400
)
return web.json_response({"error": "Both 'provider' and 'model' are required"}, status=400)
# Look up token limits from catalogue
model_info = _find_model_info(provider, model)
@@ -552,9 +546,7 @@ def _update_user_profile_memory(display_name: str, about: str) -> None:
content = build_memory_document(
name="User Profile",
description=f"User identity: {display_name}"
if display_name
else "User profile information",
description=f"User identity: {display_name}" if display_name else "User profile information",
mem_type="profile",
body=new_body if new_body else "No profile information yet.",
)
core/framework/server/routes_credentials.py
+41 -45
View File
@@ -216,9 +216,7 @@ async def handle_check_agent(request: web.Request) -> web.Response:
ensure_credential_key_env()
nodes = load_agent_nodes(agent_path)
result = validate_agent_credentials(
nodes, verify=verify, raise_on_error=False, force_refresh=True
)
result = validate_agent_credentials(nodes, verify=verify, raise_on_error=False, force_refresh=True)
# If any credential needs Aden, include ADEN_API_KEY as a first-class row
if any(c.aden_supported for c in result.credentials):
@@ -291,13 +289,15 @@ def _collect_accounts_by_provider() -> dict[str, list[dict]]:
provider = acct.get("provider", "")
if not provider:
continue
grouped.setdefault(provider, []).append({
"provider": provider,
"alias": acct.get("alias", ""),
"identity": acct.get("identity", {}) or {},
"source": acct.get("source", "aden"),
"credential_id": acct.get("credential_id", provider),
})
grouped.setdefault(provider, []).append(
{
"provider": provider,
"alias": acct.get("alias", ""),
"identity": acct.get("identity", {}) or {},
"source": acct.get("source", "aden"),
"credential_id": acct.get("credential_id", provider),
}
)
return grouped
except Exception:
logger.debug("Failed to collect accounts for specs response", exc_info=True)
@@ -327,17 +327,17 @@ async def handle_resync_credentials(request: web.Request) -> web.Response:
loop = asyncio.get_running_loop()
# _presync_aden_tokens makes blocking HTTP calls to the Aden server.
await loop.run_in_executor(
None, lambda: _presync_aden_tokens(CREDENTIAL_SPECS, force=True)
)
await loop.run_in_executor(None, lambda: _presync_aden_tokens(CREDENTIAL_SPECS, force=True))
_invalidate_queen_credentials_cache(request)
accounts_by_provider = _collect_accounts_by_provider()
return web.json_response({
"synced": True,
"accounts_by_provider": accounts_by_provider,
})
return web.json_response(
{
"synced": True,
"accounts_by_provider": accounts_by_provider,
}
)
except Exception as exc:
logger.exception("Error during credential resync: %s", exc)
return web.json_response(
@@ -366,9 +366,7 @@ async def handle_list_specs(request: web.Request) -> web.Response:
_presync_aden_tokens(CREDENTIAL_SPECS)
# Build composite store (env → encrypted file)
env_mapping = {
(spec.credential_id or name): spec.env_var for name, spec in CREDENTIAL_SPECS.items()
}
env_mapping = {(spec.credential_id or name): spec.env_var for name, spec in CREDENTIAL_SPECS.items()}
env_storage = EnvVarStorage(env_mapping=env_mapping)
if os.environ.get("HIVE_CREDENTIAL_KEY"):
storage = CompositeStorage(primary=env_storage, fallbacks=[EncryptedFileStorage()])
@@ -396,21 +394,23 @@ async def handle_list_specs(request: web.Request) -> web.Response:
available = len(accounts) > 0
else:
available = store.is_available(cred_id)
specs.append({
"credential_name": name,
"credential_id": cred_id,
"env_var": spec.env_var,
"description": spec.description,
"help_url": spec.help_url,
"api_key_instructions": spec.api_key_instructions,
"tools": spec.tools,
"aden_supported": spec.aden_supported,
"direct_api_key_supported": spec.direct_api_key_supported,
"credential_key": spec.credential_key,
"credential_group": spec.credential_group,
"available": available,
"accounts": accounts,
})
specs.append(
{
"credential_name": name,
"credential_id": cred_id,
"env_var": spec.env_var,
"description": spec.description,
"help_url": spec.help_url,
"api_key_instructions": spec.api_key_instructions,
"tools": spec.tools,
"aden_supported": spec.aden_supported,
"direct_api_key_supported": spec.direct_api_key_supported,
"credential_key": spec.credential_key,
"credential_group": spec.credential_group,
"available": available,
"accounts": accounts,
}
)
# Include aden_api_key synthetic row if any spec uses Aden
if any_aden:
@@ -422,7 +422,9 @@ async def handle_list_specs(request: web.Request) -> web.Response:
"env_var": "ADEN_API_KEY",
"description": "API key from the Developers tab in Settings",
"help_url": "https://hive.adenhq.com/",
"api_key_instructions": "1. Go to hive.adenhq.com\n2. Open Settings > Developers\n3. Copy your API key",
"api_key_instructions": (
"1. Go to hive.adenhq.com\n2. Open Settings > Developers\n3. Copy your API key"
),
"tools": [],
"aden_supported": True,
"direct_api_key_supported": True,
@@ -459,16 +461,12 @@ async def handle_validate_key(request: web.Request) -> web.Response:
api_key = body.get("api_key", "").strip()
if not provider_id or not api_key:
return web.json_response(
{"error": "provider_id and api_key are required"}, status=400
)
return web.json_response({"error": "provider_id and api_key are required"}, status=400)
try:
checker = _get_llm_key_providers().get(provider_id)
if not checker:
return web.json_response(
{"valid": True, "message": f"No health check for {provider_id}"}
)
return web.json_response({"valid": True, "message": f"No health check for {provider_id}"})
loop = asyncio.get_running_loop()
result = await loop.run_in_executor(None, lambda: checker(api_key))
@@ -476,9 +474,7 @@ async def handle_validate_key(request: web.Request) -> web.Response:
except Exception as exc:
logger.warning("LLM key validation failed for %s: %s", provider_id, exc)
return web.json_response(
{"valid": None, "message": f"Validation error: {exc}"}
)
return web.json_response({"valid": None, "message": f"Validation error: {exc}"})
def register_routes(app: web.Application) -> None:
core/framework/server/routes_events.py
+2 -6
View File
@@ -159,9 +159,7 @@ async def handle_events(request: web.Request) -> web.StreamResponse:
sse = SSEResponse()
await sse.prepare(request)
logger.info(
"SSE connected: session='%s', sub_id='%s', types=%d", session.id, sub_id, len(event_types)
)
logger.info("SSE connected: session='%s', sub_id='%s', types=%d", session.id, sub_id, len(event_types))
# Replay buffered events that were published before this SSE connected.
# The EventBus keeps a history ring-buffer; we replay the subset that
@@ -215,9 +213,7 @@ async def handle_events(request: web.Request) -> web.StreamResponse:
await sse.send_event(data)
event_count += 1
if event_count == 1:
logger.info(
"SSE first event: session='%s', type='%s'", session.id, data.get("type")
)
logger.info("SSE first event: session='%s', type='%s'", session.id, data.get("type"))
except TimeoutError:
try:
await sse.send_keepalive()
core/framework/server/routes_execution.py
+22 -53
View File
@@ -3,6 +3,7 @@
import asyncio
import json
import logging
from datetime import UTC
from typing import Any
from aiohttp import web
@@ -117,9 +118,7 @@ async def handle_trigger(request: web.Request) -> web.Response:
if session.runner:
loop = asyncio.get_running_loop()
try:
await loop.run_in_executor(
None, lambda: validate_agent_credentials(session.runner.graph.nodes)
)
await loop.run_in_executor(None, lambda: validate_agent_credentials(session.runner.graph.nodes))
except Exception as e:
agent_path = str(session.worker_path) if session.worker_path else ""
resp = _credential_error_response(e, agent_path)
@@ -129,9 +128,7 @@ async def handle_trigger(request: web.Request) -> web.Response:
# Resync MCP servers if credentials were added since the worker loaded
# (e.g. user connected an OAuth account mid-session via Aden UI).
try:
await loop.run_in_executor(
None, lambda: session.runner._tool_registry.resync_mcp_servers_if_needed()
)
await loop.run_in_executor(None, lambda: session.runner._tool_registry.resync_mcp_servers_if_needed())
except Exception as e:
logger.warning("MCP resync failed: %s", e)
@@ -228,23 +225,14 @@ async def handle_chat(request: web.Request) -> web.Response:
type(queen_executor.node_registry),
id(queen_executor.node_registry),
)
logger.debug(
"[handle_chat] node_registry keys: %s", list(queen_executor.node_registry.keys())
)
logger.debug("[handle_chat] node_registry keys: %s", list(queen_executor.node_registry.keys()))
node = queen_executor.node_registry.get("queen")
logger.debug(
"[handle_chat] node=%s, node_type=%s", node, type(node).__name__ if node else None
)
logger.debug(
"[handle_chat] has_inject_event=%s", hasattr(node, "inject_event") if node else False
)
logger.debug("[handle_chat] node=%s, node_type=%s", node, type(node).__name__ if node else None)
logger.debug("[handle_chat] has_inject_event=%s", hasattr(node, "inject_event") if node else False)
# Race condition: executor exists but node not created yet (still initializing)
if node is None and session.queen_task is not None and not session.queen_task.done():
logger.warning(
"[handle_chat] Queen executor exists but node"
" not ready yet (initializing). Waiting..."
)
logger.warning("[handle_chat] Queen executor exists but node not ready yet (initializing). Waiting...")
# Wait a short time for initialization to progress
import asyncio
@@ -302,16 +290,12 @@ async def handle_chat(request: web.Request) -> web.Response:
)
else:
logger.error(
"[handle_chat] CRITICAL: Queen node exists"
" but missing inject_event!"
" node_attrs=%s",
"[handle_chat] CRITICAL: Queen node exists but missing inject_event! node_attrs=%s",
[a for a in dir(node) if not a.startswith("_")],
)
# Queen is dead — try to revive her
logger.warning(
"[handle_chat] Queen is dead for session '%s', reviving on /chat request", session.id
)
logger.warning("[handle_chat] Queen is dead for session '%s', reviving on /chat request", session.id)
manager: Any = request.app["manager"]
try:
logger.debug("[handle_chat] Calling manager.revive_queen()...")
@@ -322,9 +306,7 @@ async def handle_chat(request: web.Request) -> web.Response:
_revived_executor = session.queen_executor
_revived_node = _revived_executor.node_registry.get("queen") if _revived_executor else None
if _revived_node is not None and hasattr(_revived_node, "inject_event"):
await _revived_node.inject_event(
message, is_client_input=True, image_content=image_content
)
await _revived_node.inject_event(message, is_client_input=True, image_content=image_content)
return web.json_response(
{
"status": "queen_revived",
@@ -552,9 +534,7 @@ async def handle_stop(request: web.Request) -> web.Response:
if hasattr(node, "cancel_current_turn"):
node.cancel_current_turn()
cancelled = await stream.cancel_execution(
execution_id, reason="Execution stopped by user"
)
cancelled = await stream.cancel_execution(execution_id, reason="Execution stopped by user")
if cancelled:
# Cancel queen's in-progress LLM turn
if session.queen_executor:
@@ -716,13 +696,12 @@ async def fork_session_into_colony(
import asyncio
import json
import shutil
from datetime import datetime, timezone
from datetime import datetime
from pathlib import Path
from framework.agent_loop.agent_loop import AgentLoop, LoopConfig
from framework.agent_loop.types import AgentContext, AgentSpec
from framework.agent_loop.types import AgentContext
from framework.server.session_manager import _queen_session_dir
from framework.storage.conversation_store import FileConversationStore
queen_loop: AgentLoop = session.queen_executor.node_registry["queen"]
queen_ctx: AgentContext = getattr(queen_loop, "_last_ctx", None)
@@ -813,11 +792,9 @@ async def fork_session_into_colony(
"queen_id": getattr(phase_state, "queen_id", "") if phase_state else "",
"loop_config": queen_lc_config,
"spawned_from": session.id,
"spawned_at": datetime.now(timezone.utc).isoformat(),
"spawned_at": datetime.now(UTC).isoformat(),
}
worker_config_path.write_text(
json.dumps(worker_meta, indent=2, ensure_ascii=False), encoding="utf-8"
)
worker_config_path.write_text(json.dumps(worker_meta, indent=2, ensure_ascii=False), encoding="utf-8")
# ── 3. Duplicate queen session into colony ───────────────────
# Copy the queen's full session directory (conversations, events,
@@ -843,9 +820,7 @@ async def fork_session_into_colony(
dest_queen_dir = _queen_session_dir(colony_session_id, queen_name)
if source_queen_dir.exists():
await asyncio.to_thread(
shutil.copytree, source_queen_dir, dest_queen_dir, dirs_exist_ok=True
)
await asyncio.to_thread(shutil.copytree, source_queen_dir, dest_queen_dir, dirs_exist_ok=True)
# Update the duplicated meta.json to point to the colony
dest_meta_path = dest_queen_dir / "meta.json"
dest_meta: dict = {}
@@ -859,9 +834,7 @@ async def fork_session_into_colony(
dest_meta["queen_id"] = queen_name
dest_meta["forked_from"] = session.id
dest_meta["colony_fork"] = True # exclude from queen DM history
dest_meta_path.write_text(
json.dumps(dest_meta, ensure_ascii=False), encoding="utf-8"
)
dest_meta_path.write_text(json.dumps(dest_meta, ensure_ascii=False), encoding="utf-8")
logger.info(
"Duplicated queen session %s -> %s for colony '%s'",
session.id,
@@ -875,9 +848,7 @@ async def fork_session_into_colony(
worker_conv_dir = worker_storage / "conversations"
source_conv_dir = dest_queen_dir / "conversations"
if source_conv_dir.exists():
await asyncio.to_thread(
shutil.copytree, source_conv_dir, worker_conv_dir, dirs_exist_ok=True
)
await asyncio.to_thread(shutil.copytree, source_conv_dir, worker_conv_dir, dirs_exist_ok=True)
logger.info("Copied queen conversations to worker storage %s", worker_conv_dir)
else:
logger.warning(
@@ -897,12 +868,12 @@ async def fork_session_into_colony(
metadata["queen_name"] = queen_name
metadata["queen_session_id"] = colony_session_id
metadata["source_session_id"] = session.id
metadata.setdefault("created_at", datetime.now(timezone.utc).isoformat())
metadata["updated_at"] = datetime.now(timezone.utc).isoformat()
metadata.setdefault("created_at", datetime.now(UTC).isoformat())
metadata["updated_at"] = datetime.now(UTC).isoformat()
metadata.setdefault("workers", {})
metadata["workers"][worker_name] = {
"task": worker_task[:100],
"spawned_at": datetime.now(timezone.utc).isoformat(),
"spawned_at": datetime.now(UTC).isoformat(),
}
metadata_path.write_text(json.dumps(metadata, indent=2, ensure_ascii=False), encoding="utf-8")
@@ -920,9 +891,7 @@ async def fork_session_into_colony(
qmeta["agent_name"] = colony_name.replace("_", " ").title()
try:
source_meta_path.parent.mkdir(parents=True, exist_ok=True)
source_meta_path.write_text(
json.dumps(qmeta, ensure_ascii=False), encoding="utf-8"
)
source_meta_path.write_text(json.dumps(qmeta, ensure_ascii=False), encoding="utf-8")
except OSError:
pass
core/framework/server/routes_queens.py
+2 -6
View File
@@ -148,9 +148,7 @@ def _transform_profile_for_api(profile: dict) -> dict:
details.append(f"Drive: {hidden['deep_motive']}")
if hidden.get("behavioral_mapping"):
details.append(f"Approach: {hidden['behavioral_mapping']}")
experience.append(
{"role": f"{profile.get('title', 'Executive Advisor')}", "details": details}
)
experience.append({"role": f"{profile.get('title', 'Executive Advisor')}", "details": details})
if experience:
result["experience"] = experience
@@ -161,9 +159,7 @@ def _transform_profile_for_api(profile: dict) -> dict:
# Signature achievement from world_lore
world_lore = profile.get("world_lore", {})
if world_lore.get("habitat"):
result["signature_achievement"] = (
f"{world_lore['habitat']}. {world_lore.get('lexicon', '')}".strip()
)
result["signature_achievement"] = f"{world_lore['habitat']}. {world_lore.get('lexicon', '')}".strip()
return result
core/framework/server/routes_sessions.py
+9 -35
View File
@@ -59,9 +59,7 @@ def _session_to_live_dict(session) -> dict:
"loaded_at": session.loaded_at,
"uptime_seconds": round(time.time() - session.loaded_at, 1),
"intro_message": getattr(session.runner, "intro_message", "") or "",
"queen_phase": phase_state.phase
if phase_state
else ("staging" if session.colony_runtime else "planning"),
"queen_phase": phase_state.phase if phase_state else ("staging" if session.colony_runtime else "planning"),
"queen_supports_images": supports_image_tool_results(queen_model) if queen_model else True,
"queen_id": getattr(phase_state, "queen_id", None) if phase_state else None,
"queen_name": (phase_state.queen_profile or {}).get("name") if phase_state else None,
@@ -229,11 +227,7 @@ async def handle_get_live_session(request: web.Request) -> web.Response:
"entry_node": ep.entry_node,
"trigger_type": ep.trigger_type,
"trigger_config": ep.trigger_config,
**(
{"next_fire_in": nf}
if (nf := rt.get_timer_next_fire_in(ep.id)) is not None
else {}
),
**({"next_fire_in": nf} if (nf := rt.get_timer_next_fire_in(ep.id)) is not None else {}),
}
for ep in rt.get_entry_points()
]
@@ -383,11 +377,7 @@ async def handle_session_entry_points(request: web.Request) -> web.Response:
"entry_node": ep.entry_node,
"trigger_type": ep.trigger_type,
"trigger_config": ep.trigger_config,
**(
{"next_fire_in": nf}
if rt and (nf := rt.get_timer_next_fire_in(ep.id)) is not None
else {}
),
**({"next_fire_in": nf} if rt and (nf := rt.get_timer_next_fire_in(ep.id)) is not None else {}),
}
for ep in eps
]
@@ -468,21 +458,13 @@ async def handle_update_trigger_task(request: web.Request) -> web.Response:
)
except ImportError:
return web.json_response(
{
"error": (
"croniter package not installed — cannot validate cron expression."
)
},
{"error": ("croniter package not installed — cannot validate cron expression.")},
status=500,
)
merged_trigger_config.pop("interval_minutes", None)
elif interval is None:
return web.json_response(
{
"error": (
"Timer trigger needs 'cron' or 'interval_minutes' in trigger_config."
)
},
{"error": ("Timer trigger needs 'cron' or 'interval_minutes' in trigger_config.")},
status=400,
)
elif not isinstance(interval, (int, float)) or interval <= 0:
@@ -580,9 +562,7 @@ async def handle_activate_trigger(request: web.Request) -> web.Response:
)
if trigger_id in getattr(session, "active_trigger_ids", set()):
return web.json_response(
{"status": "already_active", "trigger_id": trigger_id}
)
return web.json_response({"status": "already_active", "trigger_id": trigger_id})
from framework.tools.queen_lifecycle_tools import (
_persist_active_triggers,
@@ -646,9 +626,7 @@ async def handle_deactivate_trigger(request: web.Request) -> web.Response:
trigger_id = request.match_info["trigger_id"]
if trigger_id not in getattr(session, "active_trigger_ids", set()):
return web.json_response(
{"status": "already_inactive", "trigger_id": trigger_id}
)
return web.json_response({"status": "already_inactive", "trigger_id": trigger_id})
task = session.active_timer_tasks.pop(trigger_id, None)
if task and not task.done():
@@ -867,9 +845,7 @@ async def handle_delete_agent(request: web.Request) -> web.Response:
try:
shutil.rmtree(resolved)
except OSError as e:
return web.json_response(
{"error": f"Failed to delete agent directory: {e}"}, status=500
)
return web.json_response({"error": f"Failed to delete agent directory: {e}"}, status=500)
return web.json_response({"deleted": str(resolved)})
@@ -932,9 +908,7 @@ def register_routes(app: web.Application) -> None:
app.router.add_post("/api/sessions/{session_id}/reveal", handle_reveal_session_folder)
app.router.add_get("/api/sessions/{session_id}/stats", handle_session_stats)
app.router.add_get("/api/sessions/{session_id}/entry-points", handle_session_entry_points)
app.router.add_patch(
"/api/sessions/{session_id}/triggers/{trigger_id}", handle_update_trigger_task
)
app.router.add_patch("/api/sessions/{session_id}/triggers/{trigger_id}", handle_update_trigger_task)
app.router.add_post(
"/api/sessions/{session_id}/triggers/{trigger_id}/activate",
handle_activate_trigger,
core/framework/server/routes_workers.py
+4 -17
View File
@@ -70,13 +70,7 @@ async def handle_list_nodes(request: web.Request) -> web.Response:
from pathlib import Path
state_path = (
Path.home()
/ ".hive"
/ "agents"
/ session.worker_path.name
/ "sessions"
/ worker_session_id
/ "state.json"
Path.home() / ".hive" / "agents" / session.worker_path.name / "sessions" / worker_session_id / "state.json"
)
if state_path.exists():
try:
@@ -97,8 +91,7 @@ async def handle_list_nodes(request: web.Request) -> web.Response:
pass
edges = [
{"source": e.source, "target": e.target, "condition": e.condition, "priority": e.priority}
for e in graph.edges
{"source": e.source, "target": e.target, "condition": e.condition, "priority": e.priority} for e in graph.edges
]
rt = session.colony_runtime
entry_points = [
@@ -108,11 +101,7 @@ async def handle_list_nodes(request: web.Request) -> web.Response:
"entry_node": ep.entry_node,
"trigger_type": ep.trigger_type,
"trigger_config": ep.trigger_config,
**(
{"next_fire_in": nf}
if rt and (nf := rt.get_timer_next_fire_in(ep.id)) is not None
else {}
),
**({"next_fire_in": nf} if rt and (nf := rt.get_timer_next_fire_in(ep.id)) is not None else {}),
}
for ep in reg.entry_points.values()
]
@@ -250,9 +239,7 @@ async def handle_node_tools(request: web.Request) -> web.Response:
def register_routes(app: web.Application) -> None:
"""Register worker inspection routes."""
app.router.add_get("/api/sessions/{session_id}/colonies/{colony_id}/nodes", handle_list_nodes)
app.router.add_get(
"/api/sessions/{session_id}/colonies/{colony_id}/nodes/{node_id}", handle_get_node
)
app.router.add_get("/api/sessions/{session_id}/colonies/{colony_id}/nodes/{node_id}", handle_get_node)
app.router.add_get(
"/api/sessions/{session_id}/colonies/{colony_id}/nodes/{node_id}/criteria",
handle_node_criteria,
core/framework/server/session_manager.py
+15 -49
View File
@@ -120,9 +120,7 @@ class SessionManager:
(blocking I/O) then started on the event loop.
"""
def __init__(
self, model: str | None = None, credential_store=None, queen_tool_registry=None
) -> None:
def __init__(self, model: str | None = None, credential_store=None, queen_tool_registry=None) -> None:
self._sessions: dict[str, Session] = {}
self._loading: set[str] = set()
self._model = model
@@ -350,9 +348,7 @@ class SessionManager:
_colony_metadata_path = agent_path / "metadata.json"
if _colony_metadata_path.exists():
try:
_colony_metadata = json.loads(
_colony_metadata_path.read_text(encoding="utf-8")
)
_colony_metadata = json.loads(_colony_metadata_path.read_text(encoding="utf-8"))
except (json.JSONDecodeError, OSError):
pass
@@ -426,9 +422,7 @@ class SessionManager:
# Start queen with worker profile + lifecycle + monitoring tools
worker_identity = (
build_worker_profile(session.colony_runtime, agent_path=agent_path)
if session.colony_runtime
else None
build_worker_profile(session.colony_runtime, agent_path=agent_path) if session.colony_runtime else None
)
await self._start_queen(
session,
@@ -678,14 +672,10 @@ class SessionManager:
)
# Start the worker's agent loop in the background
session.queen_task = asyncio.create_task(
session.queen_executor.run(initial_message=initial_prompt)
)
session.queen_task = asyncio.create_task(session.queen_executor.run(initial_message=initial_prompt))
# Set up event persistence
if session.event_bus and queen_dir:
from framework.host.event_bus import EventBus
session.event_bus.start_persistence(queen_dir, iteration_offset=iteration_offset)
logger.info(
@@ -920,9 +910,7 @@ class SessionManager:
state.setdefault("result", {})["error"] = "Stale session: runtime restarted"
state.setdefault("timestamps", {})["updated_at"] = datetime.now().isoformat()
state_path.write_text(json.dumps(state, indent=2), encoding="utf-8")
logger.info(
"Marked stale session '%s' as cancelled for agent '%s'", d.name, agent_path.name
)
logger.info("Marked stale session '%s' as cancelled for agent '%s'", d.name, agent_path.name)
except (json.JSONDecodeError, OSError) as e:
logger.warning("Failed to clean up stale session %s: %s", d.name, e)
@@ -966,13 +954,12 @@ class SessionManager:
store = session.colony_runtime._session_store
state = await store.read_state(session_id)
if state and state.active_triggers:
from framework.host.event_bus import AgentEvent, EventType
from framework.tools.queen_lifecycle_tools import (
_start_trigger_timer,
_start_trigger_webhook,
)
from framework.host.event_bus import AgentEvent, EventType
runner = getattr(session, "runner", None)
colony_entry = runner.graph.entry_node if runner else None
saved_tasks = getattr(state, "trigger_tasks", {}) or {}
@@ -1006,11 +993,7 @@ class SessionManager:
"trigger_type": tdef.trigger_type,
"trigger_config": tdef.trigger_config,
"name": tdef.description or tdef.id,
**(
{"entry_node": colony_entry}
if colony_entry
else {}
),
**({"entry_node": colony_entry} if colony_entry else {}),
},
)
)
@@ -1059,17 +1042,13 @@ class SessionManager:
meta_path = _queen_session_dir(storage_session_id, session.queen_name) / "meta.json"
try:
_agent_name = (
session.worker_info.name
if session.worker_info
else str(agent_path.name).replace("_", " ").title()
session.worker_info.name if session.worker_info else str(agent_path.name).replace("_", " ").title()
)
existing_meta = {}
if meta_path.exists():
existing_meta = json.loads(meta_path.read_text(encoding="utf-8"))
existing_meta["agent_name"] = _agent_name
existing_meta["agent_path"] = (
str(session.worker_path) if session.worker_path else str(agent_path)
)
existing_meta["agent_path"] = str(session.worker_path) if session.worker_path else str(agent_path)
meta_path.write_text(json.dumps(existing_meta), encoding="utf-8")
except OSError:
pass
@@ -1188,9 +1167,7 @@ class SessionManager:
self._background_tasks.add(task)
task.add_done_callback(self._background_tasks.discard)
except Exception:
logger.warning(
"Session '%s': failed to spawn shutdown reflection", session_id, exc_info=True
)
logger.warning("Session '%s': failed to spawn shutdown reflection", session_id, exc_info=True)
if session.queen_task is not None:
session.queen_task.cancel()
@@ -1295,11 +1272,7 @@ class SessionManager:
_agent_name = (
session.worker_info.name
if session.worker_info
else (
str(session.worker_path.name).replace("_", " ").title()
if session.worker_path
else None
)
else (str(session.worker_path.name).replace("_", " ").title() if session.worker_path else None)
)
# Merge into existing meta.json to preserve fields written by
# _update_meta_json (e.g. phase, agent_path set during building).
@@ -1354,8 +1327,7 @@ class SessionManager:
if max_iter >= 0:
iteration_offset = max_iter + 1
logger.info(
"Session '%s' resuming with iteration_offset=%d"
" (from events.jsonl max), last phase: %s",
"Session '%s' resuming with iteration_offset=%d (from events.jsonl max), last phase: %s",
session.id,
iteration_offset,
last_phase or "unknown",
@@ -1496,8 +1468,7 @@ class SessionManager:
session.colony = colony
logger.info(
"_start_queen: unified ColonyRuntime ready for session %s "
"(%d tools, storage=%s)",
"_start_queen: unified ColonyRuntime ready for session %s (%d tools, storage=%s)",
session.id,
len(queen_tools),
queen_dir,
@@ -1529,10 +1500,7 @@ class SessionManager:
detail = cfg.get("cron") or f"every {cfg.get('interval_minutes', '?')} min"
task_info = f' -> task: "{t.task}"' if t.task else " (no task configured)"
parts.append(f" - {t.id} ({t.trigger_type}: {detail}){task_info}")
trigger_lines = (
"\n\nAvailable triggers (inactive — use set_trigger to activate):\n"
+ "\n".join(parts)
)
trigger_lines = "\n\nAvailable triggers (inactive — use set_trigger to activate):\n" + "\n".join(parts)
await node.inject_event(f"[SYSTEM] Colony loaded.{profile}{trigger_lines}")
@@ -1835,9 +1803,7 @@ class SessionManager:
if isinstance(content, list):
# Anthropic-style content blocks
content = " ".join(
b.get("text", "")
for b in content
if isinstance(b, dict) and b.get("type") == "text"
b.get("text", "") for b in content if isinstance(b, dict) and b.get("type") == "text"
)
if content and msg.get("role") == "assistant":
last_message = content[:120].strip()
core/framework/server/tests/test_api.py
+18 -53
View File
@@ -259,15 +259,11 @@ def _write_sample_session(base: Path, session_id: str):
conv_dir = session_dir / "conversations" / "node_a" / "parts"
conv_dir.mkdir(parents=True)
(conv_dir / "0001.json").write_text(json.dumps({"seq": 1, "role": "user", "content": "hello"}))
(conv_dir / "0002.json").write_text(
json.dumps({"seq": 2, "role": "assistant", "content": "hi there"})
)
(conv_dir / "0002.json").write_text(json.dumps({"seq": 2, "role": "assistant", "content": "hi there"}))
conv_dir_b = session_dir / "conversations" / "node_b" / "parts"
conv_dir_b.mkdir(parents=True)
(conv_dir_b / "0003.json").write_text(
json.dumps({"seq": 3, "role": "user", "content": "continue"})
)
(conv_dir_b / "0003.json").write_text(json.dumps({"seq": 3, "role": "user", "content": "continue"}))
# Logs
logs_dir = session_dir / "logs"
@@ -291,9 +287,7 @@ def _write_sample_session(base: Path, session_id: str):
"attention_reasons": ["retried"],
"total_steps": 1,
}
(logs_dir / "details.jsonl").write_text(
json.dumps(detail_a) + "\n" + json.dumps(detail_b) + "\n"
)
(logs_dir / "details.jsonl").write_text(json.dumps(detail_a) + "\n" + json.dumps(detail_b) + "\n")
step_a = {"node_id": "node_a", "step_index": 0, "llm_text": "thinking..."}
step_b = {"node_id": "node_b", "step_index": 0, "llm_text": "retrying..."}
@@ -302,9 +296,7 @@ def _write_sample_session(base: Path, session_id: str):
return session_id, session_dir, state
def _write_queen_session(
tmp_path: Path, queen_id: str, session_id: str, meta: dict | None = None
) -> Path:
def _write_queen_session(tmp_path: Path, queen_id: str, session_id: str, meta: dict | None = None) -> Path:
"""Create a persisted queen session directory for restore tests."""
session_dir = tmp_path / ".hive" / "agents" / "queens" / queen_id / "sessions" / session_id
session_dir.mkdir(parents=True)
@@ -394,9 +386,7 @@ class TestSessionCRUD:
async def test_create_session_with_worker_forwards_session_id(self):
app = create_app()
manager = app["manager"]
manager.create_session_with_worker_colony = AsyncMock(
return_value=_make_session(agent_id="my-custom-session")
)
manager.create_session_with_worker_colony = AsyncMock(return_value=_make_session(agent_id="my-custom-session"))
async with TestClient(TestServer(app)) as client:
resp = await client.post(
@@ -600,14 +590,10 @@ class TestMessageBootstrap:
manager.build_llm = MagicMock(return_value=MagicMock())
manager.stop_session = AsyncMock()
manager.create_session = AsyncMock()
monkeypatch.setattr(
routes_messages, "select_queen", AsyncMock(return_value="queen_technology")
)
monkeypatch.setattr(routes_messages, "select_queen", AsyncMock(return_value="queen_technology"))
async with TestClient(TestServer(app)) as client:
resp = await client.post(
"/api/messages/classify", json={"message": "Build me a scraper"}
)
resp = await client.post("/api/messages/classify", json={"message": "Build me a scraper"})
assert resp.status == 200
data = await resp.json()
# Assert inside the async-with so app shutdown (which stops
@@ -623,9 +609,7 @@ class TestQueenSessionSelection:
@pytest.mark.asyncio
async def test_select_queen_session_rejects_foreign_session(self, monkeypatch, tmp_path):
_patch_queen_storage(monkeypatch, tmp_path)
_write_queen_session(
tmp_path, "queen_growth", "other_session", {"queen_id": "queen_growth"}
)
_write_queen_session(tmp_path, "queen_growth", "other_session", {"queen_id": "queen_growth"})
app = create_app()
async with TestClient(TestServer(app)) as client:
@@ -663,9 +647,7 @@ class TestQueenSessionSelection:
assert any(call.args == ("other_live",) for call in manager.stop_session.await_args_list)
@pytest.mark.asyncio
async def test_select_queen_session_restores_specific_history_session(
self, monkeypatch, tmp_path
):
async def test_select_queen_session_restores_specific_history_session(self, monkeypatch, tmp_path):
_patch_queen_storage(monkeypatch, tmp_path)
_write_queen_session(
tmp_path,
@@ -1167,9 +1149,7 @@ class TestGraphNodes:
assert data["entry_node"] == "node_a"
@pytest.mark.asyncio
async def test_list_nodes_with_session_enrichment(
self, nodes_and_edges, sample_session, tmp_agent_dir
):
async def test_list_nodes_with_session_enrichment(self, nodes_and_edges, sample_session, tmp_agent_dir):
session_id, session_dir, state = sample_session
tmp_path, agent_name, base = tmp_agent_dir
nodes, edges = nodes_and_edges
@@ -1182,9 +1162,7 @@ class TestGraphNodes:
app = _make_app_with_session(session)
async with TestClient(TestServer(app)) as client:
resp = await client.get(
f"/api/sessions/test_agent/graphs/primary/nodes?session_id={session_id}"
)
resp = await client.get(f"/api/sessions/test_agent/graphs/primary/nodes?session_id={session_id}")
assert resp.status == 200
data = await resp.json()
node_map = {n["id"]: n for n in data["nodes"]}
@@ -1233,9 +1211,7 @@ class TestGraphNodes:
assert resp.status == 200
data = await resp.json()
assert "system_prompt" in data
assert (
data["system_prompt"] == "You are a helpful assistant that produces valid results."
)
assert data["system_prompt"] == "You are a helpful assistant that produces valid results."
# Node without system_prompt should return empty string
resp2 = await client.get("/api/sessions/test_agent/graphs/primary/nodes/node_b")
@@ -1270,9 +1246,7 @@ class TestNodeCriteria:
assert data["output_keys"] == ["result"]
@pytest.mark.asyncio
async def test_criteria_with_log_enrichment(
self, nodes_and_edges, sample_session, tmp_agent_dir
):
async def test_criteria_with_log_enrichment(self, nodes_and_edges, sample_session, tmp_agent_dir):
"""Criteria endpoint enriched with last execution from logs."""
session_id, session_dir, state = sample_session
tmp_path, agent_name, base = tmp_agent_dir
@@ -1293,8 +1267,7 @@ class TestNodeCriteria:
async with TestClient(TestServer(app)) as client:
resp = await client.get(
f"/api/sessions/test_agent/graphs/primary/nodes/node_b/criteria"
f"?session_id={session_id}"
f"/api/sessions/test_agent/graphs/primary/nodes/node_b/criteria?session_id={session_id}"
)
assert resp.status == 200
data = await resp.json()
@@ -1311,9 +1284,7 @@ class TestNodeCriteria:
app = _make_app_with_session(session)
async with TestClient(TestServer(app)) as client:
resp = await client.get(
"/api/sessions/test_agent/graphs/primary/nodes/nonexistent/criteria"
)
resp = await client.get("/api/sessions/test_agent/graphs/primary/nodes/nonexistent/criteria")
assert resp.status == 404
@@ -1388,9 +1359,7 @@ class TestLogs:
app = _make_app_with_session(session)
async with TestClient(TestServer(app)) as client:
resp = await client.get(
f"/api/sessions/test_agent/logs?session_id={session_id}&level=summary"
)
resp = await client.get(f"/api/sessions/test_agent/logs?session_id={session_id}&level=summary")
assert resp.status == 200
data = await resp.json()
assert data["run_id"] == session_id
@@ -1411,9 +1380,7 @@ class TestLogs:
app = _make_app_with_session(session)
async with TestClient(TestServer(app)) as client:
resp = await client.get(
f"/api/sessions/test_agent/logs?session_id={session_id}&level=details"
)
resp = await client.get(f"/api/sessions/test_agent/logs?session_id={session_id}&level=details")
assert resp.status == 200
data = await resp.json()
assert data["session_id"] == session_id
@@ -1435,9 +1402,7 @@ class TestLogs:
app = _make_app_with_session(session)
async with TestClient(TestServer(app)) as client:
resp = await client.get(
f"/api/sessions/test_agent/logs?session_id={session_id}&level=tools"
)
resp = await client.get(f"/api/sessions/test_agent/logs?session_id={session_id}&level=tools")
assert resp.status == 200
data = await resp.json()
assert data["session_id"] == session_id
core/framework/skills/catalog.py
+10 -4
View File
@@ -26,12 +26,15 @@ Before replying: scan <available_skills> <description> entries.
- If multiple could apply: choose the most specific one, then read/follow it.
- If none clearly apply: do not read any SKILL.md.
Constraints: never read more than one skill up front; only read after selecting.
- When a skill drives external API writes (Gmail, Calendar, GitHub, etc.), assume rate limits: prefer fewer larger writes, avoid tight one-item loops, serialize bursts when possible, and respect 429/Retry-After.
- When a skill drives external API writes (Gmail, Calendar, GitHub, etc.),
assume rate limits: prefer fewer larger writes, avoid tight one-item loops,
serialize bursts when possible, and respect 429/Retry-After.
The following skills provide specialized instructions for specific tasks.
Use `read_file` to load a skill's SKILL.md when the task matches its description.
When a skill file references a relative path, resolve it against the skill directory (parent of SKILL.md) and use that absolute path in tool commands."""
When a skill file references a relative path, resolve it against the
skill directory (parent of SKILL.md) and use that absolute path in tool commands."""
_MANDATORY_HEADER_COMPACT = """## Skills (mandatory)
Before replying: scan <available_skills> <name> entries.
@@ -39,12 +42,15 @@ Before replying: scan <available_skills> <name> entries.
- If multiple could apply: choose the most specific one, then read/follow it.
- If none clearly apply: do not read any SKILL.md.
Constraints: never read more than one skill up front; only read after selecting.
- When a skill drives external API writes (Gmail, Calendar, GitHub, etc.), assume rate limits: prefer fewer larger writes, avoid tight one-item loops, serialize bursts when possible, and respect 429/Retry-After.
- When a skill drives external API writes (Gmail, Calendar, GitHub, etc.),
assume rate limits: prefer fewer larger writes, avoid tight one-item loops,
serialize bursts when possible, and respect 429/Retry-After.
The following skills provide specialized instructions for specific tasks.
Use `read_file` to load a skill's SKILL.md when the task matches its name.
When a skill file references a relative path, resolve it against the skill directory (parent of SKILL.md) and use that absolute path in tool commands."""
When a skill file references a relative path, resolve it against the
skill directory (parent of SKILL.md) and use that absolute path in tool commands."""
class SkillCatalog:
core/framework/skills/cli.py
+9 -26
View File
@@ -134,9 +134,7 @@ def register_skill_commands(subparsers) -> None:
info_parser.set_defaults(func=cmd_skill_info)
# hive skill init
init_parser = skill_sub.add_parser(
"init", help="Scaffold a new skill directory with a SKILL.md template"
)
init_parser = skill_sub.add_parser("init", help="Scaffold a new skill directory with a SKILL.md template")
init_parser.add_argument("--name", dest="skill_name", default=None, metavar="NAME")
init_parser.add_argument(
"--dir",
@@ -193,9 +191,7 @@ def register_skill_commands(subparsers) -> None:
update_parser.set_defaults(func=cmd_skill_update)
# hive skill search
search_parser = skill_sub.add_parser(
"search", help="Search the skill registry by name, tag, or description"
)
search_parser = skill_sub.add_parser("search", help="Search the skill registry by name, tag, or description")
search_parser.add_argument("query", help="Search query string")
search_parser.add_argument("--json", action="store_true", help="Output as JSON")
search_parser.set_defaults(func=cmd_skill_search)
@@ -231,9 +227,7 @@ def register_skill_commands(subparsers) -> None:
fork_parser.set_defaults(func=cmd_skill_fork)
# hive skill test
test_parser = skill_sub.add_parser(
"test", help="Run a skill in isolation or execute its eval suite (CLI-9)"
)
test_parser = skill_sub.add_parser("test", help="Run a skill in isolation or execute its eval suite (CLI-9)")
test_parser.add_argument("path", help="Path to SKILL.md or its parent directory")
test_parser.add_argument(
"--input",
@@ -649,9 +643,7 @@ def cmd_skill_validate(args) -> int:
print(f"✓ {path} — valid ({len(result.warnings)} warning(s))")
return 0
else:
print(
f"✗ {path} — invalid ({len(result.errors)} error(s), {len(result.warnings)} warning(s))"
)
print(f"✗ {path} — invalid ({len(result.errors)} error(s), {len(result.warnings)} warning(s))")
return 1
@@ -672,9 +664,7 @@ def cmd_skill_doctor(args) -> int:
for skill_name, dir_name in SKILL_REGISTRY.items():
skill_md = _DEFAULT_SKILLS_DIR / dir_name / "SKILL.md"
if use_json:
report = _doctor_skill_file(
skill_name, skill_md, parse_skill_md, json_mode=True, scope="framework"
)
report = _doctor_skill_file(skill_name, skill_md, parse_skill_md, json_mode=True, scope="framework")
overall_errors += len(report["errors"])
skill_results.append(report)
else:
@@ -696,9 +686,7 @@ def cmd_skill_doctor(args) -> int:
candidate = USER_SKILLS_DIR / args.name / "SKILL.md"
if candidate.exists():
if use_json:
report = _doctor_skill_file(
args.name, candidate, parse_skill_md, json_mode=True, scope="user"
)
report = _doctor_skill_file(args.name, candidate, parse_skill_md, json_mode=True, scope="user")
print(_json.dumps({"skills": [report], "total_errors": len(report["errors"])}))
return 1 if report["errors"] else 0
print(f"\nChecking skill: {args.name} [user]")
@@ -838,8 +826,7 @@ def cmd_skill_update(args) -> int:
if not installed_version and not use_json:
print(
f"Warning: installed skill '{args.name}' has no version field — "
"cannot compare. Re-installing.",
f"Warning: installed skill '{args.name}' has no version field — cannot compare. Re-installing.",
file=sys.stderr,
)
@@ -1057,9 +1044,7 @@ def cmd_skill_test(args) -> int:
# ── 4. Structural-only mode (no LLM needed) ───────────────────────────────
if not has_input and not has_evals:
doctor_errors = _doctor_skill_file(
skill.name, path, parse_skill_md, json_mode=use_json, scope="user"
)
doctor_errors = _doctor_skill_file(skill.name, path, parse_skill_md, json_mode=use_json, scope="user")
if use_json:
print(
_json.dumps(
@@ -1209,9 +1194,7 @@ def cmd_skill_test(args) -> int:
constraint=assertion,
source_document=eval_prompt,
summary=skill_response,
criteria=(
"Evaluate whether the skill response satisfies the assertion."
),
criteria=("Evaluate whether the skill response satisfies the assertion."),
)
passes = judged.get("passes", False)
explanation = judged.get("explanation", "")
core/framework/skills/defaults.py
+2 -5
View File
@@ -60,9 +60,7 @@ def _apply_overrides(skill_name: str, body: str, overrides: dict[str, Any]) -> s
# Convert float warn_at_usage_ratio → warn_at_usage_ratio_pct for the placeholder
if "warn_at_usage_ratio" in overrides:
overrides = dict(overrides)
overrides.setdefault(
"warn_at_usage_ratio_pct", int(float(overrides["warn_at_usage_ratio"]) * 100)
)
overrides.setdefault("warn_at_usage_ratio_pct", int(float(overrides["warn_at_usage_ratio"]) * 100))
values = {**defaults, **overrides}
for key, val in values.items():
body = body.replace(f"{{{{{key}}}}}", str(val))
@@ -189,8 +187,7 @@ class DefaultSkillManager:
approx_tokens = len(combined) // 4
if approx_tokens > 2000:
logger.warning(
"Default skill protocols exceed 2000 token budget "
"(~%d tokens, %d chars). Consider trimming.",
"Default skill protocols exceed 2000 token budget (~%d tokens, %d chars). Consider trimming.",
approx_tokens,
len(combined),
)
core/framework/skills/registry.py
+1 -2
View File
@@ -23,8 +23,7 @@ logger = logging.getLogger(__name__)
# Default registry index URL (Phase 3 repo, may not exist yet)
_DEFAULT_REGISTRY_URL = (
"https://raw.githubusercontent.com/hive-skill-registry/"
"hive-skill-registry/main/skill_index.json"
"https://raw.githubusercontent.com/hive-skill-registry/hive-skill-registry/main/skill_index.json"
)
_CACHE_DIR = Path.home() / ".hive" / "registry_cache"
core/framework/skills/skill_errors.py
+1 -3
View File
@@ -34,9 +34,7 @@ class SkillError(Exception):
self.what = what
self.why = why
self.fix = fix
self.message = (
f"[{self.code.value}]\nWhat failed: {self.what}\nWhy: {self.why}\nFix: {self.fix}"
)
self.message = f"[{self.code.value}]\nWhat failed: {self.what}\nWhy: {self.why}\nFix: {self.fix}"
super().__init__(self.message)
core/framework/skills/tool_gating.py
+1 -1
View File
@@ -15,8 +15,8 @@ descriptions to get picked up on demand.
from __future__ import annotations
import logging
from collections.abc import Iterable
from pathlib import Path
from typing import Iterable
logger = logging.getLogger(__name__)
core/framework/skills/trust.py
+3 -10
View File
@@ -413,8 +413,7 @@ class TrustGate:
return
self._print("")
self._print(
f"{Colors.YELLOW}Security notice:{Colors.NC} Skills inject instructions "
"into the agent's system prompt."
f"{Colors.YELLOW}Security notice:{Colors.NC} Skills inject instructions into the agent's system prompt."
)
self._print(
" Only load skills from sources you trust. "
@@ -441,10 +440,7 @@ class TrustGate:
p(f"{Colors.YELLOW}{'=' * 60}{Colors.NC}")
p("")
proj_label = str(project_dir) if project_dir else "this project"
p(
f" The project at {Colors.CYAN}{proj_label}{Colors.NC} wants to load "
f"{len(project_skills)} skill(s)"
)
p(f" The project at {Colors.CYAN}{proj_label}{Colors.NC} wants to load {len(project_skills)} skill(s)")
p(" that will inject instructions into the agent's system prompt.")
if repo_key:
p(f" Source: {Colors.BOLD}{repo_key}{Colors.NC}")
@@ -458,10 +454,7 @@ class TrustGate:
p(" Options:")
p(f" {Colors.CYAN}1){Colors.NC} Trust this session only")
p(f" {Colors.CYAN}2){Colors.NC} Trust permanently — remember for future runs")
p(
f" {Colors.DIM}3) Deny"
f" — skip all project-scope skills from this repo{Colors.NC}"
)
p(f" {Colors.DIM}3) Deny — skip all project-scope skills from this repo{Colors.NC}")
p(f"{Colors.YELLOW}{'─' * 60}{Colors.NC}")
def _prompt_consent(self, Colors) -> str: # noqa: N803
core/framework/skills/validator.py
+5 -15
View File
@@ -82,8 +82,7 @@ def validate_strict(path: Path) -> ValidationResult:
frontmatter = yaml.safe_load(raw_yaml)
except yaml.YAMLError as exc:
errors.append(
f"YAML parse error: {exc}. "
'Wrap values containing colons in quotes, e.g. description: "Use for: research".'
f'YAML parse error: {exc}. Wrap values containing colons in quotes, e.g. description: "Use for: research".'
)
return ValidationResult(passed=False, errors=errors, warnings=warnings)
@@ -101,10 +100,7 @@ def validate_strict(path: Path) -> ValidationResult:
# 6. name present and non-empty (no directory-name fallback in strict mode)
name = frontmatter.get("name")
if not name or not str(name).strip():
errors.append(
"Missing required field: 'name' must be present. "
"Add 'name: your-skill-name' to the frontmatter."
)
errors.append("Missing required field: 'name' must be present. Add 'name: your-skill-name' to the frontmatter.")
else:
name = str(name).strip()
parent_dir_name = path.parent.name
@@ -112,8 +108,7 @@ def validate_strict(path: Path) -> ValidationResult:
# 7. name length <= 64 chars
if len(name) > _MAX_NAME_LENGTH:
errors.append(
f"Skill name '{name}' is {len(name)} characters — "
f"maximum is {_MAX_NAME_LENGTH}. Shorten the name."
f"Skill name '{name}' is {len(name)} characters — maximum is {_MAX_NAME_LENGTH}. Shorten the name."
)
# 8. name matches parent directory (dot-namespace prefix allowed: hive.X with dir X)
@@ -125,10 +120,7 @@ def validate_strict(path: Path) -> ValidationResult:
# 9. body non-empty
if not body:
errors.append(
"Skill body (instructions) is empty. "
"Add markdown instructions after the closing --- delimiter."
)
errors.append("Skill body (instructions) is empty. Add markdown instructions after the closing --- delimiter.")
# 10. license present — warning only
if not frontmatter.get("license"):
@@ -142,9 +134,7 @@ def validate_strict(path: Path) -> ValidationResult:
for script_path in sorted(scripts_dir.iterdir()):
if script_path.is_file():
if not (script_path.stat().st_mode & (stat.S_IXUSR | stat.S_IXGRP | stat.S_IXOTH)):
errors.append(
f"Script not executable: {script_path.name}. Run: chmod +x {script_path}"
)
errors.append(f"Script not executable: {script_path.name}. Run: chmod +x {script_path}")
# 12. allowed-tools entries are non-empty strings — warning if malformed
allowed_tools = frontmatter.get("allowed-tools")
core/framework/storage/checkpoint_store.py
+2 -6
View File
@@ -123,9 +123,7 @@ class CheckpointStore:
return None
try:
return CheckpointIndex.model_validate_json(
self.index_path.read_text(encoding="utf-8")
)
return CheckpointIndex.model_validate_json(self.index_path.read_text(encoding="utf-8"))
except Exception as e:
logger.error(f"Failed to load checkpoint index: {e}")
return None
@@ -317,9 +315,7 @@ class CheckpointStore:
# Update latest_checkpoint_id if we removed the latest
if index.latest_checkpoint_id == checkpoint_id:
index.latest_checkpoint_id = (
index.checkpoints[-1].checkpoint_id if index.checkpoints else None
)
index.latest_checkpoint_id = index.checkpoints[-1].checkpoint_id if index.checkpoints else None
# Write updated index
await asyncio.to_thread(_write, index)
core/framework/testing/approval_cli.py
+4 -14
View File
@@ -96,20 +96,14 @@ def batch_approval(
# Validate request
valid, error = req.validate_action()
if not valid:
results.append(
ApprovalResult.error_result(req.test_id, req.action, error or "Invalid request")
)
results.append(ApprovalResult.error_result(req.test_id, req.action, error or "Invalid request"))
counts["errors"] += 1
continue
# Load test
test = storage.load_test(goal_id, req.test_id)
if not test:
results.append(
ApprovalResult.error_result(
req.test_id, req.action, f"Test {req.test_id} not found"
)
)
results.append(ApprovalResult.error_result(req.test_id, req.action, f"Test {req.test_id} not found"))
counts["errors"] += 1
continue
@@ -132,9 +126,7 @@ def batch_approval(
storage.update_test(test)
results.append(
ApprovalResult.success_result(
req.test_id, req.action, f"Test {req.action.value}d successfully"
)
ApprovalResult.success_result(req.test_id, req.action, f"Test {req.action.value}d successfully")
)
except Exception as e:
@@ -233,9 +225,7 @@ def _process_action(
test.approve()
storage.update_test(test)
print("✓ Approved (no modifications)")
return ApprovalResult.success_result(
test.id, ApprovalAction.APPROVE, "No modifications made"
)
return ApprovalResult.success_result(test.id, ApprovalAction.APPROVE, "No modifications made")
elif action == ApprovalAction.SKIP:
print("⏭ Skipped (remains pending)")
core/framework/testing/approval_types.py
+1 -3
View File
@@ -61,9 +61,7 @@ class ApprovalResult(BaseModel):
timestamp: datetime = Field(default_factory=datetime.now)
@classmethod
def success_result(
cls, test_id: str, action: ApprovalAction, message: str | None = None
) -> "ApprovalResult":
def success_result(cls, test_id: str, action: ApprovalAction, message: str | None = None) -> "ApprovalResult":
"""Create a successful result."""
return cls(
test_id=test_id,
core/framework/testing/categorizer.py
+4 -10
View File
@@ -81,9 +81,7 @@ class ErrorCategorizer:
def __init__(self):
"""Initialize categorizer with compiled patterns."""
self._logic_patterns = [re.compile(p, re.IGNORECASE) for p in self.LOGIC_ERROR_PATTERNS]
self._impl_patterns = [
re.compile(p, re.IGNORECASE) for p in self.IMPLEMENTATION_ERROR_PATTERNS
]
self._impl_patterns = [re.compile(p, re.IGNORECASE) for p in self.IMPLEMENTATION_ERROR_PATTERNS]
self._edge_patterns = [re.compile(p, re.IGNORECASE) for p in self.EDGE_CASE_PATTERNS]
def categorize(self, result: TestResult) -> ErrorCategory | None:
@@ -192,8 +190,7 @@ class ErrorCategorizer:
"The goal specification may not accurately describe the desired behavior."
),
ErrorCategory.IMPLEMENTATION_ERROR: (
"Fix the code in agent nodes/edges. "
"There's a bug in the implementation that needs to be corrected."
"Fix the code in agent nodes/edges. There's a bug in the implementation that needs to be corrected."
),
ErrorCategory.EDGE_CASE: (
"Add a new test for this edge case scenario. "
@@ -226,17 +223,14 @@ class ErrorCategorizer:
"action": "Fix nodes/edges implementation",
"restart_required": False,
"description": (
"There's a code bug. Fix the agent implementation, "
"then re-run Eval (skip Goal stage)."
"There's a code bug. Fix the agent implementation, then re-run Eval (skip Goal stage)."
),
},
ErrorCategory.EDGE_CASE: {
"stage": "Eval",
"action": "Add new test only",
"restart_required": False,
"description": (
"This is a new scenario. Add a test for it and continue in the Eval stage."
),
"description": ("This is a new scenario. Add a test for it and continue in the Eval stage."),
},
}
return guidance.get(
core/framework/testing/debug_tool.py
+3 -6
View File
@@ -244,12 +244,10 @@ class DebugTool:
return {
"execution_path": run.metrics.nodes_executed if hasattr(run, "metrics") else [],
"decisions": [
d.model_dump() if hasattr(d, "model_dump") else str(d)
for d in getattr(run, "decisions", [])
d.model_dump() if hasattr(d, "model_dump") else str(d) for d in getattr(run, "decisions", [])
],
"problems": [
p.model_dump() if hasattr(p, "model_dump") else str(p)
for p in getattr(run, "problems", [])
p.model_dump() if hasattr(p, "model_dump") else str(p) for p in getattr(run, "problems", [])
],
"status": run.status.value if hasattr(run, "status") else "unknown",
}
@@ -284,8 +282,7 @@ class DebugTool:
if failures_by_category["uncategorized"]:
suggestions.append(
f"Found {len(failures_by_category['uncategorized'])} uncategorized failures. "
"Manual review required."
f"Found {len(failures_by_category['uncategorized'])} uncategorized failures. Manual review required."
)
return suggestions

Some files were not shown because too many files have changed in this diff Show More