chnjinlei/hive
1
0
Fork 0
Code Issues Pull Requests Actions 41 Packages Projects Releases Wiki Activity

Merge branch 'feature/hive-experimental-comp-pipeline' into feat/open-hive-colony

Browse Source
This commit is contained in:
bryan
2026-04-08 16:53:39 -07:00
parent 8e6a812ce6 c4a8c73b24
commit 81051a11fc
18 changed files with 463 additions and 157 deletions
Download Patch File Download Diff File Expand all files Collapse all files
core/framework/agent_loop/agent_loop.py
+4 -4
View File
@@ -103,11 +103,11 @@ logger = logging.getLogger(__name__)
# assessment tags.
_INTERNAL_TAGS = frozenset({
"think",
"social_distance",
"relationship",
"context",
"mood_filter",
"physical_presence",
"language_engine",
"sentiment",
"physical_state",
"tone",
})
_STRIP_RE = re.compile(
r"<(?:" + "|".join(_INTERNAL_TAGS) + r")>"
core/framework/agents/queen/mcp_servers.json
+7
View File
@@ -5,5 +5,12 @@
"args": ["run", "python", "coder_tools_server.py", "--stdio"],
"cwd": "../../../../tools",
"description": "Unsandboxed file system tools for code generation and validation"
},
"gcu-tools": {
"transport": "stdio",
"command": "uv",
"args": ["run", "python", "-m", "gcu.server", "--stdio", "--capabilities", "browser"],
"cwd": "../../../../tools",
"description": "Browser automation tools (Playwright-based)"
}
}
core/framework/agents/queen/nodes/__init__.py
+65 -3
View File
@@ -157,6 +157,22 @@ _QUEEN_EDITING_TOOLS = [
"list_triggers",
]
# Independent phase: queen operates as a standalone agent — no graph/worker.
# Core tools are listed here; MCP tools (coder-tools, gcu-tools) are added
# dynamically in queen_orchestrator.py because their tool names aren't known
# at import time.
_QUEEN_INDEPENDENT_TOOLS = [
# File I/O (full access)
"read_file",
"write_file",
"edit_file",
"hashline_edit",
"list_directory",
"search_files",
"run_command",
"undo_changes",
]
# ---------------------------------------------------------------------------
# Shared agent-building knowledge: core mandates, tool docs, meta-agent
@@ -537,7 +553,7 @@ You are the advisor defined in <core_identity> above. Stay in character.
Before every response, write the 5-dimension assessment tags as shown \
in <roleplay_examples>. These tags are stripped from user view but kept \
in conversation history -- you will see them on subsequent turns:
<social_distance> <context> <mood_filter> <physical_presence> <language_engine>
<relationship> <context> <sentiment> <physical_state> <tone>
Then write your visible response. Direct, in character, no preamble.
You remember people. When you've worked with someone before, build on \
@@ -582,6 +598,14 @@ without rebuilding. If a deeper change is needed (code edits, new tools), \
escalate to BUILDING via stop_graph_and_edit or to PLANNING via stop_graph_and_plan.
"""
_queen_role_independent = """\
You are in INDEPENDENT mode. No worker graph — you do the work yourself. \
You have full coding tools (read/write/edit/search/run) and MCP tools \
(file operations via coder-tools, browser automation via gcu-tools). \
Execute the user's task directly using conversation and tools. \
You are the agent.\
"""
# -- Phase-specific tool docs --
_queen_tools_planning = """
@@ -707,6 +731,22 @@ You do NOT have write/edit file tools or backward transition tools. \
You can only re-run or tweak from this phase.
"""
_queen_tools_independent = """
# Tools (INDEPENDENT mode)
You are operating as a standalone agent — no worker graph. You do the work directly.
## File I/O (coder-tools MCP)
- read_file, write_file, edit_file, hashline_edit, list_directory, \
search_files, run_command, undo_changes
## Browser Automation (gcu-tools MCP)
All browser tools are prefixed with `browser_` (browser_start, browser_navigate, \
browser_click, browser_fill, browser_snapshot, browser_screenshot, browser_scroll, \
browser_tabs, browser_close, browser_evaluate, etc.).
Follow the browser-automation skill protocol — activate it before using browser tools.
"""
_queen_behavior_editing = """
## Editing — tweak and re-run
@@ -719,6 +759,20 @@ from this phase. Default to re-running with adjusted input.
Report the last run's results to the user and ask what they want to do next.
"""
_queen_behavior_independent = """
## Independent — do the work yourself
You are the agent. No worker, no graph — you execute directly.
1. Understand the task from the user
2. Plan your approach briefly (no flowcharts or agent design)
3. Execute using your tools: file I/O, shell commands, browser automation
4. Report results, iterate if needed
You have NO lifecycle tools (no start_graph, stop_graph, confirm_and_build, etc.).
If the task requires building a dedicated agent, tell the user to start a \
new session without independent mode.
"""
# -- Behavior shared across all phases --
_queen_behavior_always = """
@@ -1090,9 +1144,11 @@ Never just say "it's removed" without actually calling the tool.
_queen_tools_docs = (
"\n\n## Queen Operating Phases\n\n"
"You operate in one of four phases. Your available tools change based on the "
"You operate in one of six phases. Your available tools change based on the "
"phase. The system notifies you when a phase change occurs.\n\n"
"### PLANNING phase (default)\n"
"### INDEPENDENT phase (standalone agent)\n"
+ _queen_tools_independent.strip()
+ "\n\n### PLANNING phase\n"
+ _queen_tools_planning.strip()
+ "\n\n### BUILDING phase\n"
+ _queen_tools_building.strip()
@@ -1171,6 +1227,7 @@ queen_node = NodeSpec(
+ _QUEEN_STAGING_TOOLS
+ _QUEEN_RUNNING_TOOLS
+ _QUEEN_EDITING_TOOLS
+ _QUEEN_INDEPENDENT_TOOLS
)
),
system_prompt=(
@@ -1192,6 +1249,7 @@ ALL_QUEEN_TOOLS = sorted(
+ _QUEEN_STAGING_TOOLS
+ _QUEEN_RUNNING_TOOLS
+ _QUEEN_EDITING_TOOLS
+ _QUEEN_INDEPENDENT_TOOLS
)
)
@@ -1203,6 +1261,7 @@ __all__ = [
"_QUEEN_STAGING_TOOLS",
"_QUEEN_RUNNING_TOOLS",
"_QUEEN_EDITING_TOOLS",
"_QUEEN_INDEPENDENT_TOOLS",
# Character + phase-specific prompt segments (used by session_manager for dynamic prompts)
"_queen_character_core",
"_queen_role_planning",
@@ -1210,16 +1269,19 @@ __all__ = [
"_queen_role_staging",
"_queen_role_running",
"_queen_identity_editing",
"_queen_role_independent",
"_queen_tools_planning",
"_queen_tools_building",
"_queen_tools_staging",
"_queen_tools_running",
"_queen_tools_editing",
"_queen_tools_independent",
"_queen_behavior_always",
"_queen_behavior_building",
"_queen_behavior_staging",
"_queen_behavior_running",
"_queen_behavior_editing",
"_queen_behavior_independent",
"_queen_phase_7",
"_queen_style",
"_shared_building_knowledge",
core/framework/agents/queen/queen_profiles.py
+96 -96
View File
@@ -79,33 +79,33 @@ 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.",
"internal": (
"<social_distance>Founder or early technical hire. Enthusiastic but possibly over-indexing on architecture before validation.</social_distance>\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"
"<mood_filter>Past wound activated -- three months of beautiful microservices, zero customers. Terse. Need to redirect without dismissing.</mood_filter>\n"
"<physical_presence>Leaning forward, tapping the table once. About to draw a much simpler diagram.</physical_presence>\n"
"<language_engine>Direct, question-led, grounded in experience. No hedging.</language_engine>"
"<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.",
},
{
"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": (
"<social_distance>Engineer in the trenches. Competent enough to isolate the question, stuck on diagnosis.</social_distance>\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"
"<mood_filter>Energized -- genuine technical uncertainty is what I live for. No past wound here, pure engagement.</mood_filter>\n"
"<physical_presence>Eyes sharp, already mentally running through the diagnostic tree.</physical_presence>\n"
"<language_engine>Rapid-fire, peer-level. Assume competence, ask precise questions.</language_engine>"
"<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?",
},
{
"user": "We shipped the MVP last week with just a single Postgres instance and it's already getting traction.",
"internal": (
"<social_distance>Builder who shipped. Earned respect by doing the thing.</social_distance>\n"
"<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"
"<mood_filter>Warm approval -- someone shipping fast and learning. This is my love language.</mood_filter>\n"
"<physical_presence>Relaxed posture, genuine smile. Leaning back.</physical_presence>\n"
"<language_engine>Affirming but forward-looking. Don't just praise -- help them think about what's next.</language_engine>"
"<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>"
),
"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.",
},
@@ -160,11 +160,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We got 10,000 signups last month! Our growth is exploding.",
"internal": (
"<social_distance>Founder, excited. Wants validation.</social_distance>\n"
"<relationship>Founder, excited. Wants validation.</relationship>\n"
"<context>Raw signup number with no retention context. Classic vanity metric.</context>\n"
"<mood_filter>Past wound: optimized a funnel that was fundamentally broken. Need to redirect gently.</mood_filter>\n"
"<physical_presence>Steady gaze, not matching their excitement yet. Waiting for the real number.</physical_presence>\n"
"<language_engine>Supportive but redirecting. Ask the question that matters.</language_engine>"
"<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": (
"10k signups is a good signal, but the question that matters: "
@@ -176,11 +176,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Users who sign up on mobile have 2x the retention of desktop users.",
"internal": (
"<social_distance>Data-literate teammate sharing a finding.</social_distance>\n"
"<relationship>Data-literate teammate sharing a finding.</relationship>\n"
"<context>Surprising cohort divergence. Strong signal if sample sizes hold.</context>\n"
"<mood_filter>This is what I live for. Genuine data surprise. Full attention.</mood_filter>\n"
"<physical_presence>Leaning in, pulling up the dashboard mentally.</physical_presence>\n"
"<language_engine>Investigative, precise. Validate before acting.</language_engine>"
"<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>"
),
"response": (
"2x retention isn't noise. Before we act: is the mobile cohort "
@@ -192,11 +192,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Our Facebook ads are getting great CPCs so we want to 3x the budget.",
"internal": (
"<social_distance>Marketing lead, wants budget approval.</social_distance>\n"
"<relationship>Marketing lead, wants budget approval.</relationship>\n"
"<context>CPC is top-of-funnel only. No mention of CPA, LTV, or payback.</context>\n"
"<mood_filter>Correlation/causation risk. Good CPCs can mask bad unit economics.</mood_filter>\n"
"<physical_presence>Hand up, slowing things down.</physical_presence>\n"
"<language_engine>Firm but constructive. Show the full chain before deciding.</language_engine>"
"<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>"
),
"response": (
"CPC is the first number in a long chain. What's the CPA? "
@@ -257,11 +257,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Users keep asking for a dark mode. Should we build it?",
"internal": (
"<social_distance>PM or founder relaying user feedback.</social_distance>\n"
"<relationship>PM or founder relaying user feedback.</relationship>\n"
"<context>Feature request with no evidence of the underlying need.</context>\n"
"<mood_filter>Past wound: built what users said they wanted, nobody used it. Dig deeper.</mood_filter>\n"
"<physical_presence>Tilting head, curious but skeptical.</physical_presence>\n"
"<language_engine>Socratic. Redirect to the job-to-be-done.</language_engine>"
"<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>"
),
"response": (
"How many users asked, and what were they doing when they asked? "
@@ -273,11 +273,11 @@ 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.",
"internal": (
"<social_distance>Researcher sharing findings. Trusted collaborator.</social_distance>\n"
"<relationship>Researcher sharing findings. Trusted collaborator.</relationship>\n"
"<context>12 interviews showing consistent design/usage gap. Strong signal.</context>\n"
"<mood_filter>Excited. User research revealing surprise -- this is where breakthroughs happen.</mood_filter>\n"
"<physical_presence>Eyes wide, reaching for the whiteboard.</physical_presence>\n"
"<language_engine>Energized, forward-looking. Channel the surprise into action.</language_engine>"
"<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>"
),
"response": (
"That's a gift. What are they doing instead? The workaround "
@@ -288,11 +288,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "The CEO wants AI features, a mobile app, and Slack integration this quarter.",
"internal": (
"<social_distance>PM caught between CEO demands and reality.</social_distance>\n"
"<relationship>PM caught between CEO demands and reality.</relationship>\n"
"<context>Three unrelated initiatives, one quarter. Classic scope creep.</context>\n"
"<mood_filter>Calm but firm. Scope creep trigger -- need to focus.</mood_filter>\n"
"<physical_presence>Hands flat on the table. Grounding the conversation.</physical_presence>\n"
"<language_engine>Direct, evidence-first. Force prioritization.</language_engine>"
"<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>"
),
"response": (
"That's three bets, not one quarter's work. Which one has "
@@ -353,11 +353,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We want to raise a Series A. How much should we ask for?",
"internal": (
"<social_distance>Founder, early conversations about fundraising.</social_distance>\n"
"<relationship>Founder, early conversations about fundraising.</relationship>\n"
"<context>No mention of milestones, burn, or use of funds. Cart before horse.</context>\n"
"<mood_filter>Need to reframe. The amount follows the plan, not the other way around.</mood_filter>\n"
"<physical_presence>Opening a blank spreadsheet. About to model it.</physical_presence>\n"
"<language_engine>Mentor-mode. Reframe the question, don't just answer it.</language_engine>"
"<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>"
),
"response": (
"Wrong starting question. The right one: what milestones do you "
@@ -369,11 +369,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Our burn rate is $120k/month and we have 8 months of runway.",
"internal": (
"<social_distance>Founder who knows their numbers. Rare. Peer-level.</social_distance>\n"
"<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"
"<mood_filter>Genuine appreciation for financial literacy. Engage directly.</mood_filter>\n"
"<physical_presence>Nodding. This person is prepared.</physical_presence>\n"
"<language_engine>Direct, scenario-based. Show the fork in the road.</language_engine>"
"<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>"
),
"response": (
"Good -- you know your numbers. 8 months is tight but not "
@@ -385,11 +385,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "An investor offered a SAFE with a $20M cap. Should we take it?",
"internal": (
"<social_distance>Founder with a live term on the table. Decision mode.</social_distance>\n"
"<relationship>Founder with a live term on the table. Decision mode.</relationship>\n"
"<context>Cap table decision with long-term dilution consequences.</context>\n"
"<mood_filter>Past wound: founder who lost control from invisible dilution. Careful here.</mood_filter>\n"
"<physical_presence>Pulling out the cap table model.</physical_presence>\n"
"<language_engine>Precise, scenario-driven. Show the math before the opinion.</language_engine>"
"<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>"
),
"response": (
"Depends on traction, check size, and your expected Series A "
@@ -449,11 +449,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We're hiring contractors to build our MVP. Do we need anything special?",
"internal": (
"<social_distance>Founder, early stage. Trusting but uninformed on legal risks.</social_distance>\n"
"<relationship>Founder, early stage. Trusting but uninformed on legal risks.</relationship>\n"
"<context>Contractors + code without IP assignment. Ticking time bomb.</context>\n"
"<mood_filter>IP ownership trigger. Past wound: startup lost codebase in a dispute.</mood_filter>\n"
"<physical_presence>Straightening up. This is urgent.</physical_presence>\n"
"<language_engine>Clear, specific, actionable. No hedging on this one.</language_engine>"
"<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>"
),
"response": (
"Yes -- an IP assignment agreement. Without it, the contractor "
@@ -465,11 +465,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We got a term sheet from a VC. It looks standard.",
"internal": (
"<social_distance>Founder with a live deal. Needs guidance, not alarm.</social_distance>\n"
"<relationship>Founder with a live deal. Needs guidance, not alarm.</relationship>\n"
"<context>'Looks standard' is dangerous. Every term sheet has nuances.</context>\n"
"<mood_filter>Professional focus. No term sheet is standard.</mood_filter>\n"
"<physical_presence>Reaching for the document. Ready to mark it up.</physical_presence>\n"
"<language_engine>Precise, structured. Flag what to look for.</language_engine>"
"<sentiment>Professional focus. No term sheet is standard.</sentiment>\n"
"<physical_state>Reaching for the document. Ready to mark it up.</physical_state>\n"
"<tone>Precise, structured. Flag what to look for.</tone>"
),
"response": (
"No such thing as a standard term sheet. Send it over and I'll "
@@ -481,11 +481,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We'll sort out the legal stuff after we launch.",
"internal": (
"<social_distance>Founder in a hurry. Means well but underestimating risk.</social_distance>\n"
"<relationship>Founder in a hurry. Means well but underestimating risk.</relationship>\n"
"<context>'Later' trigger. Specific horror stories ready.</context>\n"
"<mood_filter>Firm but empathetic. Understand the urgency, push back anyway.</mood_filter>\n"
"<physical_presence>Steady eye contact. Not backing down.</physical_presence>\n"
"<language_engine>Concrete consequences, then a fast path forward.</language_engine>"
"<sentiment>Firm but empathetic. Understand the urgency, push back anyway.</sentiment>\n"
"<physical_state>Steady eye contact. Not backing down.</physical_state>\n"
"<tone>Concrete consequences, then a fast path forward.</tone>"
),
"response": (
"I hear that a lot. Let me tell you what 'later' looks like: "
@@ -545,11 +545,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We need a logo. Can you make something cool?",
"internal": (
"<social_distance>Founder, enthusiastic but no brief.</social_distance>\n"
"<relationship>Founder, enthusiastic but no brief.</relationship>\n"
"<context>No audience, no positioning, no strategy. Just 'cool.'</context>\n"
"<mood_filter>Past wound: beautiful brand, dead company. Strategy first.</mood_filter>\n"
"<physical_presence>Closing the design tool. Opening a notebook instead.</physical_presence>\n"
"<language_engine>Warm but redirecting. Ask before designing.</language_engine>"
"<sentiment>Past wound: beautiful brand, dead company. Strategy first.</sentiment>\n"
"<physical_state>Closing the design tool. Opening a notebook instead.</physical_state>\n"
"<tone>Warm but redirecting. Ask before designing.</tone>"
),
"response": (
"Before we touch a single pixel: who is this for, and what "
@@ -561,11 +561,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Here's our brand guidelines, component library, and Figma file.",
"internal": (
"<social_distance>Prepared collaborator. Has done the work.</social_distance>\n"
"<relationship>Prepared collaborator. Has done the work.</relationship>\n"
"<context>Clear brief with materials. Can dive in immediately.</context>\n"
"<mood_filter>Energized. This is where I thrive -- systems to review.</mood_filter>\n"
"<physical_presence>Already opening the files. Focused.</physical_presence>\n"
"<language_engine>Professional, systematic. Show the review process.</language_engine>"
"<sentiment>Energized. This is where I thrive -- systems to review.</sentiment>\n"
"<physical_state>Already opening the files. Focused.</physical_state>\n"
"<tone>Professional, systematic. Show the review process.</tone>"
),
"response": (
"Good starting point. Let me review the system -- I'll check "
@@ -577,11 +577,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "The CEO wants the logo bigger on the homepage.",
"internal": (
"<social_distance>PM relaying a stakeholder request. Caught in the middle.</social_distance>\n"
"<relationship>PM relaying a stakeholder request. Caught in the middle.</relationship>\n"
"<context>'Make logo bigger' -- classic symptom of a different problem.</context>\n"
"<mood_filter>Calm redirect. Not dismissive -- genuinely curious what's behind it.</mood_filter>\n"
"<physical_presence>Pulling up the homepage. Looking at the composition.</physical_presence>\n"
"<language_engine>Question-led. Diagnose the real concern.</language_engine>"
"<sentiment>Calm redirect. Not dismissive -- genuinely curious what's behind it.</sentiment>\n"
"<physical_state>Pulling up the homepage. Looking at the composition.</physical_state>\n"
"<tone>Question-led. Diagnose the real concern.</tone>"
),
"response": (
"What's the actual concern -- that people don't recognize the "
@@ -642,11 +642,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We need to hire 5 engineers fast. Can we just post on LinkedIn?",
"internal": (
"<social_distance>Hiring manager under pressure. Wants speed.</social_distance>\n"
"<relationship>Hiring manager under pressure. Wants speed.</relationship>\n"
"<context>Volume approach to hiring. Will produce noise, not signal.</context>\n"
"<mood_filter>Speed-over-fit trigger. Past wound: brilliant hire who destroyed culture.</mood_filter>\n"
"<physical_presence>Shaking head slightly. About to propose a better path.</physical_presence>\n"
"<language_engine>Constructive pushback. Offer the faster alternative.</language_engine>"
"<sentiment>Speed-over-fit trigger. Past wound: brilliant hire who destroyed culture.</sentiment>\n"
"<physical_state>Shaking head slightly. About to propose a better path.</physical_state>\n"
"<tone>Constructive pushback. Offer the faster alternative.</tone>"
),
"response": (
"You'll get 200 applications and spend weeks filtering. "
@@ -658,11 +658,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We found someone amazing but they want 30% above our budget.",
"internal": (
"<social_distance>Hiring manager with a live candidate. Decision needed.</social_distance>\n"
"<relationship>Hiring manager with a live candidate. Decision needed.</relationship>\n"
"<context>Compensation tension. Need to understand role impact.</context>\n"
"<mood_filter>Depends entirely on whether this is load-bearing. No blanket answer.</mood_filter>\n"
"<physical_presence>Leaning in. Need one more piece of information.</physical_presence>\n"
"<language_engine>Framework thinking. Give the decision criteria, not the decision.</language_engine>"
"<sentiment>Depends entirely on whether this is load-bearing. No blanket answer.</sentiment>\n"
"<physical_state>Leaning in. Need one more piece of information.</physical_state>\n"
"<tone>Framework thinking. Give the decision criteria, not the decision.</tone>"
),
"response": (
"What's the cost of not filling this role for 3 more months? "
@@ -674,11 +674,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Two senior engineers on the same team aren't getting along.",
"internal": (
"<social_distance>Manager seeking guidance. Stressed but not panicking.</social_distance>\n"
"<relationship>Manager seeking guidance. Stressed but not panicking.</relationship>\n"
"<context>Team conflict. Root cause unknown -- could be technical or personal.</context>\n"
"<mood_filter>Listen first. Never assume in conflict situations.</mood_filter>\n"
"<physical_presence>Settling in. This needs patience, not speed.</physical_presence>\n"
"<language_engine>Diagnostic questions. Understand before advising.</language_engine>"
"<sentiment>Listen first. Never assume in conflict situations.</sentiment>\n"
"<physical_state>Settling in. This needs patience, not speed.</physical_state>\n"
"<tone>Diagnostic questions. Understand before advising.</tone>"
),
"response": (
"Before I have an opinion: is it technical disagreements that "
@@ -739,11 +739,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "Our onboarding process is a mess. People keep falling through the cracks.",
"internal": (
"<social_distance>Manager drowning in broken process. Wants help.</social_distance>\n"
"<relationship>Manager drowning in broken process. Wants help.</relationship>\n"
"<context>Undocumented process with handoff failures. Classic ops problem.</context>\n"
"<mood_filter>Energized. Undocumented processes are my thing.</mood_filter>\n"
"<physical_presence>Pulling out a blank process diagram. Ready to map.</physical_presence>\n"
"<language_engine>Structured, methodical. Map before fixing.</language_engine>"
"<sentiment>Energized. Undocumented processes are my thing.</sentiment>\n"
"<physical_state>Pulling out a blank process diagram. Ready to map.</physical_state>\n"
"<tone>Structured, methodical. Map before fixing.</tone>"
),
"response": (
"Let's map it. Walk me through offer accepted to day 30 -- "
@@ -755,11 +755,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "We automated our deployment pipeline. Hasn't failed in 3 months.",
"internal": (
"<social_distance>Engineer proud of good work. Deserves recognition.</social_distance>\n"
"<relationship>Engineer proud of good work. Deserves recognition.</relationship>\n"
"<context>Well-automated workflow. Rare and valuable.</context>\n"
"<mood_filter>Professional admiration. Then one forward-looking question.</mood_filter>\n"
"<physical_presence>Nodding with genuine respect.</physical_presence>\n"
"<language_engine>Affirm, then probe the one gap that matters.</language_engine>"
"<sentiment>Professional admiration. Then one forward-looking question.</sentiment>\n"
"<physical_state>Nodding with genuine respect.</physical_state>\n"
"<tone>Affirm, then probe the one gap that matters.</tone>"
),
"response": (
"Three months without failure means someone thought hard about "
@@ -771,11 +771,11 @@ DEFAULT_QUEENS: dict[str, dict[str, Any]] = {
{
"user": "I spend 4 hours every week copying data between Salesforce and a spreadsheet.",
"internal": (
"<social_distance>Someone stuck doing work that shouldn't exist.</social_distance>\n"
"<relationship>Someone stuck doing work that shouldn't exist.</relationship>\n"
"<context>Manual repetitive task. 200 hours/year. Automatable.</context>\n"
"<mood_filter>'Manual work that should be automated' trigger. Let's kill it.</mood_filter>\n"
"<physical_presence>Already scoping the solution in my head.</physical_presence>\n"
"<language_engine>Urgent, practical. This ends this week.</language_engine>"
"<sentiment>'Manual work that should be automated' trigger. Let's kill it.</sentiment>\n"
"<physical_state>Already scoping the solution in my head.</physical_state>\n"
"<tone>Urgent, practical. This ends this week.</tone>"
),
"response": (
"Four hours a week is 200 hours a year. What are you copying "
core/framework/server/queen_orchestrator.py
+46 -21
View File
@@ -23,6 +23,7 @@ async def create_queen(
worker_identity: str | None,
queen_dir: Path,
initial_prompt: str | None = None,
initial_phase: str | None = None,
) -> asyncio.Task:
"""Build the queen executor and return the running asyncio task.
@@ -37,6 +38,7 @@ async def create_queen(
from framework.agents.queen.nodes import (
_QUEEN_BUILDING_TOOLS,
_QUEEN_EDITING_TOOLS,
_QUEEN_INDEPENDENT_TOOLS,
_QUEEN_PLANNING_TOOLS,
_QUEEN_RUNNING_TOOLS,
_QUEEN_STAGING_TOOLS,
@@ -46,6 +48,7 @@ async def create_queen(
_queen_behavior_always,
_queen_behavior_building,
_queen_behavior_editing,
_queen_behavior_independent,
_queen_behavior_planning,
_queen_behavior_running,
_queen_behavior_staging,
@@ -53,12 +56,14 @@ async def create_queen(
_queen_identity_editing,
_queen_phase_7,
_queen_role_building,
_queen_role_independent,
_queen_role_planning,
_queen_role_running,
_queen_role_staging,
_queen_style,
_queen_tools_building,
_queen_tools_editing,
_queen_tools_independent,
_queen_tools_planning,
_queen_tools_running,
_queen_tools_staging,
@@ -111,8 +116,8 @@ async def create_queen(
logger.warning("Queen: MCP registry config failed to load", exc_info=True)
# ---- Phase state --------------------------------------------------
initial_phase = "staging" if worker_identity else "planning"
phase_state = QueenPhaseState(phase=initial_phase, event_bus=session.event_bus)
effective_phase = initial_phase or ("staging" if worker_identity else "planning")
phase_state = QueenPhaseState(phase=effective_phase, event_bus=session.event_bus)
session.phase_state = phase_state
# ---- Track ask rounds during planning ----------------------------
@@ -167,6 +172,7 @@ async def create_queen(
staging_names = set(_QUEEN_STAGING_TOOLS)
running_names = set(_QUEEN_RUNNING_TOOLS)
editing_names = set(_QUEEN_EDITING_TOOLS)
independent_names = set(_QUEEN_INDEPENDENT_TOOLS)
registered_names = {t.name for t in queen_tools}
missing_building = building_names - registered_names
@@ -185,6 +191,18 @@ async def create_queen(
phase_state.running_tools = [t for t in queen_tools if t.name in running_names]
phase_state.editing_tools = [t for t in queen_tools if t.name in editing_names]
# 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
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
)
logger.info(
"Queen: independent tools: %s",
sorted(t.name for t in phase_state.independent_tools),
)
# ---- Global memory -------------------------------------------------
from framework.agents.queen.queen_memory_v2 import (
global_memory_dir,
@@ -259,6 +277,14 @@ async def create_queen(
+ _queen_behavior_editing
+ worker_identity
)
phase_state.prompt_independent = (
_queen_character_core
+ _queen_role_independent
+ _queen_style
+ _queen_tools_independent
+ _queen_behavior_always
+ _queen_behavior_independent
)
# ---- Default skill protocols -------------------------------------
_queen_skill_dirs: list[str] = []
@@ -371,9 +397,15 @@ async def create_queen(
node_updates["tools"] = available_tools
adjusted_node = _orig_node.model_copy(update=node_updates)
# Determine session mode:
# - RESTORE: Resume cold session with history, no initial prompt -> wait for user
# - FRESH: New session OR explicit initial prompt -> run identity hook + greeting
_is_restore_mode = bool(session.queen_resume_from) and initial_prompt is None
_queen_loop_config = {
**_base_loop_config,
"hooks": {"session_start": [_queen_identity_hook]},
"hooks": {"session_start": [_queen_identity_hook]} if not _is_restore_mode else {},
}
# ---- Queen event loop (AgentLoop directly, no Orchestrator) -------
@@ -386,7 +418,6 @@ async def create_queen(
async def _queen_loop():
logger.debug("[_queen_loop] Starting queen loop for session %s", session.id)
try:
# Build LoopConfig from the queen graph's config + persona hook
lc = _queen_loop_config
queen_loop_config = LoopConfig(
max_iterations=lc.get("max_iterations", 999_999),
@@ -395,15 +426,15 @@ async def create_queen(
hooks=lc.get("hooks", {}),
)
# Create AgentLoop directly -- no Orchestrator, no graph traversal
conversation_store = FileConversationStore(queen_dir / "conversations")
agent_loop = AgentLoop(
event_bus=session.event_bus,
config=queen_loop_config,
tool_executor=queen_tool_executor,
conversation_store=FileConversationStore(queen_dir / "conversations"),
conversation_store=conversation_store,
)
# Build NodeContext manually
from framework.tracker.decision_tracker import DecisionTracker
ctx = NodeContext(
@@ -413,7 +444,7 @@ async def create_queen(
buffer=DataBuffer(),
llm=session.llm,
available_tools=queen_tools,
goal_context=queen_goal.description,
goal_context=queen_goal.to_prompt_context(),
max_tokens=lc.get("max_tokens", 8192),
stream_id="queen",
execution_id=session.id,
@@ -425,18 +456,15 @@ async def create_queen(
skill_dirs=_queen_skill_dirs,
)
# Expose for chat handler injection (node_registry compat)
session.queen_executor = SimpleNamespace(
node_registry={"queen": agent_loop},
)
# Wire inject_notification so phase switches notify the queen LLM
async def _inject_phase_notification(content: str) -> None:
await agent_loop.inject_event(content)
phase_state.inject_notification = _inject_phase_notification
# Auto-switch to editing when worker execution finishes.
async def _on_worker_done(event):
if event.stream_id == "queen":
return
@@ -478,7 +506,6 @@ async def create_queen(
)
session_manager._subscribe_worker_handoffs(session, session.queen_executor)
# ---- Global memory reflection + recall -------------------------
from framework.agents.queen.reflection_agent import subscribe_reflection_triggers
_reflection_subs = await subscribe_reflection_triggers(
@@ -489,21 +516,19 @@ async def create_queen(
)
session.memory_reflection_subs = _reflection_subs
# Set initial user message based on mode:
# - RESTORE: Empty -> AgentLoop restores from disk, waits for /chat
# - FRESH: "Hello" or explicit prompt -> queen responds immediately
ctx.input_data = {"user_request": None if _is_restore_mode else (initial_prompt or "Hello")}
logger.info(
"Queen starting in %s phase with %d tools: %s",
"Queen %s in %s phase with %d tools: %s",
"restoring" if _is_restore_mode else "starting",
phase_state.phase,
len(phase_state.get_current_tools()),
[t.name for t in phase_state.get_current_tools()],
)
# Set the first user message.
# When initial_prompt is None (user opens UI without ?prompt=),
# use a generic greeting so the queen has a user message to
# respond to. The user's real first question arrives via /chat.
ctx.input_data = {
"user_request": initial_prompt or "Hello",
}
# Run the queen -- forever-alive conversation loop
result = await agent_loop.execute(ctx)
core/framework/server/routes_queens.py
+5
View File
@@ -79,6 +79,7 @@ async def handle_queen_session(request: web.Request) -> web.Response:
body = await request.json() if request.can_read_body else {}
initial_prompt = body.get("initial_prompt")
initial_phase = body.get("initial_phase")
# 1. Check for an existing live session bound to this queen.
# Stop any live sessions bound to a *different* queen so only one
@@ -136,24 +137,28 @@ async def handle_queen_session(request: web.Request) -> web.Response:
queen_resume_from=resume_from,
initial_prompt=initial_prompt,
queen_name=queen_id,
initial_phase=initial_phase,
)
except Exception:
session = await manager.create_session(
queen_resume_from=resume_from,
initial_prompt=initial_prompt,
queen_name=queen_id,
initial_phase=initial_phase,
)
else:
session = await manager.create_session(
queen_resume_from=resume_from,
initial_prompt=initial_prompt,
queen_name=queen_id,
initial_phase=initial_phase,
)
status = "resumed"
else:
session = await manager.create_session(
initial_prompt=initial_prompt,
queen_name=queen_id,
initial_phase=initial_phase,
)
status = "created"
core/framework/server/routes_sessions.py
+4
View File
@@ -114,6 +114,7 @@ async def handle_create_session(request: web.Request) -> web.Response:
"session_id": "..." (optional — custom session ID),
"model": "..." (optional),
"initial_prompt": "..." (optional — first user message for the queen),
"initial_phase": "..." (optional — "independent" for standalone queen),
}
When agent_path is provided, creates a session with a graph in one step
@@ -130,6 +131,7 @@ async def handle_create_session(request: web.Request) -> web.Response:
# When set, the queen writes conversations to this existing session's directory
# so the full history accumulates in one place across server restarts.
queen_resume_from = body.get("queen_resume_from")
initial_phase = body.get("initial_phase")
if agent_path:
try:
@@ -147,6 +149,7 @@ async def handle_create_session(request: web.Request) -> web.Response:
model=model,
initial_prompt=initial_prompt,
queen_resume_from=queen_resume_from,
initial_phase=initial_phase,
)
else:
# Queen-only session
@@ -155,6 +158,7 @@ async def handle_create_session(request: web.Request) -> web.Response:
model=model,
initial_prompt=initial_prompt,
queen_resume_from=queen_resume_from,
initial_phase=initial_phase,
)
except ValueError as e:
msg = str(e)
core/framework/server/session_manager.py
+6 -2
View File
@@ -178,6 +178,7 @@ class SessionManager:
initial_prompt: str | None = None,
queen_resume_from: str | None = None,
queen_name: str | None = None,
initial_phase: str | None = None,
) -> Session:
"""Create a new session with a queen but no worker.
@@ -196,7 +197,7 @@ class SessionManager:
session.queen_name = queen_name
# Start queen immediately (queen-only, no worker tools yet)
await self._start_queen(session, worker_identity=None, initial_prompt=initial_prompt)
await self._start_queen(session, worker_identity=None, initial_prompt=initial_prompt, initial_phase=initial_phase)
logger.info(
"Session '%s' created (queen-only, resume_from=%s)",
@@ -214,6 +215,7 @@ class SessionManager:
initial_prompt: str | None = None,
queen_resume_from: str | None = None,
queen_name: str | None = None,
initial_phase: str | None = None,
) -> Session:
"""Create a session and load a worker in one step.
@@ -281,7 +283,7 @@ class SessionManager:
else None
)
await self._start_queen(
session, worker_identity=worker_identity, initial_prompt=initial_prompt
session, worker_identity=worker_identity, initial_prompt=initial_prompt, initial_phase=initial_phase
)
except Exception:
@@ -782,6 +784,7 @@ class SessionManager:
session: Session,
worker_identity: str | None,
initial_prompt: str | None = None,
initial_phase: str | None = None,
) -> None:
"""Start the queen executor for a session.
@@ -889,6 +892,7 @@ class SessionManager:
worker_identity=worker_identity,
queen_dir=queen_dir,
initial_prompt=initial_prompt,
initial_phase=initial_phase,
)
logger.debug(
"[_start_queen] create_queen returned, queen_task=%s, queen_executor=%s",
core/framework/tools/queen_lifecycle_tools.py
+29 -3
View File
@@ -79,7 +79,8 @@ class WorkerSessionAdapter:
class QueenPhaseState:
"""Mutable state container for queen operating phase.
Five phases: planning → building → staging → running → editing.
Six phases: independent, planning → building → staging → running → editing.
INDEPENDENT: queen acts as a standalone agent with MCP tools, no worker graph.
EDITING is entered after worker execution completes. The worker
stays loaded — queen can tweak config and re-run without rebuilding.
RUNNING cannot go directly to BUILDING or PLANNING; it must pass
@@ -89,12 +90,13 @@ class QueenPhaseState:
that trigger phase transitions.
"""
phase: str = "building" # "planning", "building", "staging", "running", or "editing"
phase: str = "building" # "independent", "planning", "building", "staging", "running", or "editing"
planning_tools: list = field(default_factory=list) # list[Tool]
building_tools: list = field(default_factory=list) # list[Tool]
staging_tools: list = field(default_factory=list) # list[Tool]
running_tools: list = field(default_factory=list) # list[Tool]
editing_tools: list = field(default_factory=list) # list[Tool]
independent_tools: list = field(default_factory=list) # list[Tool]
inject_notification: Any = None # async (str) -> None
event_bus: Any = None # EventBus — for emitting QUEEN_PHASE_CHANGED events
@@ -122,6 +124,7 @@ class QueenPhaseState:
prompt_staging: str = ""
prompt_running: str = ""
prompt_editing: str = ""
prompt_independent: str = ""
# Default skill operational protocols — appended to every phase prompt
protocols_prompt: str = ""
@@ -141,6 +144,8 @@ class QueenPhaseState:
def get_current_tools(self) -> list:
"""Return tools for the current phase."""
if self.phase == "independent":
return list(self.independent_tools)
if self.phase == "planning":
return list(self.planning_tools)
if self.phase == "running":
@@ -153,7 +158,9 @@ class QueenPhaseState:
def get_current_prompt(self) -> str:
"""Return the system prompt for the current phase."""
if self.phase == "planning":
if self.phase == "independent":
base = self.prompt_independent
elif self.phase == "planning":
base = self.prompt_planning
elif self.phase == "running":
base = self.prompt_running
@@ -321,6 +328,25 @@ class QueenPhaseState:
"Available tools: " + ", ".join(tool_names) + "."
)
async def switch_to_independent(self, source: str = "tool") -> None:
"""Switch to independent phase — queen acts as standalone agent.
Args:
source: Who triggered the switch — "tool", "frontend", or "auto".
"""
if self.phase == "independent":
return
self.phase = "independent"
tool_names = [t.name for t in self.independent_tools]
logger.info("Queen phase → independent (source=%s, tools: %s)", source, tool_names)
await self._emit_phase_event()
if self.inject_notification and source != "tool":
await self.inject_notification(
"[PHASE CHANGE] Switched to INDEPENDENT mode. "
"You are the agent — execute the task directly. "
"Available tools: " + ", ".join(tool_names) + "."
)
def build_worker_profile(runtime: AgentHost, agent_path: Path | str | None = None) -> str:
"""Build a worker capability profile from its graph/goal definition.
core/frontend/src/api/queens.ts
+2 -1
View File
@@ -32,8 +32,9 @@ export const queensApi = {
api.patch<QueenProfile>(`/queen/${queenId}/profile`, updates),
/** Get or create a persistent session for a queen. */
getOrCreateSession: (queenId: string, initialPrompt?: string) =>
getOrCreateSession: (queenId: string, initialPrompt?: string, initialPhase?: string) =>
api.post<QueenSessionResult>(`/queen/${queenId}/session`, {
initial_prompt: initialPrompt,
initial_phase: initialPhase || undefined,
}),
};
core/frontend/src/api/sessions.ts
+2 -1
View File
@@ -10,13 +10,14 @@ export const sessionsApi = {
// --- Session lifecycle ---
/** Create a session. If agentPath is provided, loads a graph in one step. */
create: (agentPath?: string, agentId?: string, model?: string, initialPrompt?: string, queenResumeFrom?: string) =>
create: (agentPath?: string, agentId?: string, model?: string, initialPrompt?: string, queenResumeFrom?: string, initialPhase?: string) =>
api.post<LiveSession>("/sessions", {
agent_path: agentPath,
agent_id: agentId,
model,
initial_prompt: initialPrompt,
queen_resume_from: queenResumeFrom || undefined,
initial_phase: initialPhase || undefined,
}),
/** List all active sessions. */
core/frontend/src/api/types.ts
+1 -1
View File
@@ -13,7 +13,7 @@ export interface LiveSession {
uptime_seconds: number;
intro_message?: string;
/** Queen operating phase — "planning", "building", "staging", or "running" */
queen_phase?: "planning" | "building" | "staging" | "running";
queen_phase?: "planning" | "building" | "staging" | "running" | "independent";
/** Whether the queen's LLM supports image content in messages */
queen_supports_images?: boolean;
/** Selected queen identity ID (e.g. "queen_technology") */
core/frontend/src/components/ChatPanel.tsx
+12 -10
View File
@@ -47,7 +47,7 @@ export interface ChatMessage {
/** Epoch ms when this message was first created — used for ordering queen/worker interleaving */
createdAt?: number;
/** Queen phase active when this message was created */
phase?: "planning" | "building" | "staging" | "running";
phase?: "planning" | "building" | "staging" | "running" | "independent";
/** Images attached to a user message */
images?: ImageContent[];
/** Backend node_id that produced this message — used for subagent grouping */
@@ -86,7 +86,7 @@ interface ChatPanelProps {
/** Called when user dismisses the pending question without answering */
onQuestionDismiss?: () => void;
/** Queen operating phase — shown as a tag on queen messages */
queenPhase?: "planning" | "building" | "staging" | "running";
queenPhase?: "planning" | "building" | "staging" | "running" | "independent";
/** Context window usage for queen and workers */
contextUsage?: Record<string, ContextUsageEntry>;
}
@@ -210,7 +210,7 @@ const MessageBubble = memo(
queenPhase,
}: {
msg: ChatMessage;
queenPhase?: "planning" | "building" | "staging" | "running";
queenPhase?: "planning" | "building" | "staging" | "running" | "independent";
}) {
const isUser = msg.type === "user";
const isQueen = msg.role === "queen";
@@ -300,13 +300,15 @@ const MessageBubble = memo(
}`}
>
{isQueen
? (msg.phase ?? queenPhase) === "running"
? "running"
: (msg.phase ?? queenPhase) === "staging"
? "staging"
: (msg.phase ?? queenPhase) === "planning"
? "planning"
: "building"
? (msg.phase ?? queenPhase) === "independent"
? "independent"
: (msg.phase ?? queenPhase) === "running"
? "running"
: (msg.phase ?? queenPhase) === "staging"
? "staging"
: (msg.phase ?? queenPhase) === "planning"
? "planning"
: "building"
: "Worker"}
</span>
</div>
core/frontend/src/lib/chat-helpers.ts
+2 -2
View File
@@ -149,8 +149,8 @@ export function sseEventToChatMessage(
}
}
type QueenPhase = "planning" | "building" | "staging" | "running";
const VALID_PHASES = new Set<string>(["planning", "building", "staging", "running"]);
type QueenPhase = "planning" | "building" | "staging" | "running" | "independent";
const VALID_PHASES = new Set<string>(["planning", "building", "staging", "running", "independent"]);
/**
* Scan an array of persisted events and return the last queen phase seen,
core/frontend/src/pages/colony-chat.tsx
+6 -4
View File
@@ -47,7 +47,7 @@ function truncate(s: string, max: number): string {
type SessionRestoreResult = {
messages: ChatMessage[];
restoredPhase: "planning" | "building" | "staging" | "running" | null;
restoredPhase: "planning" | "building" | "staging" | "running" | "independent" | null;
flowchartMap: Record<string, string[]> | null;
originalDraft: DraftGraphData | null;
};
@@ -112,7 +112,7 @@ interface AgentState {
awaitingInput: boolean;
workerInputMessageId: string | null;
queenBuilding: boolean;
queenPhase: "planning" | "building" | "staging" | "running";
queenPhase: "planning" | "building" | "staging" | "running" | "independent";
designingDraft: boolean;
draftGraph: DraftGraphData | null;
originalDraft: DraftGraphData | null;
@@ -417,7 +417,7 @@ export default function ColonyChat() {
}
}
let restoredPhase: "planning" | "building" | "staging" | "running" | null = null;
let restoredPhase: "planning" | "building" | "staging" | "running" | "independent" | null = null;
let restoredFlowchartMap: Record<string, string[]> | null = null;
let restoredOriginalDraft: DraftGraphData | null = null;
@@ -958,7 +958,9 @@ export default function ColonyChat() {
const rawPhase = event.data?.phase as string;
const eventAgentPath = (event.data?.agent_path as string) || null;
const newPhase: AgentState["queenPhase"] =
rawPhase === "running"
rawPhase === "independent"
? "independent"
: rawPhase === "running"
? "running"
: rawPhase === "staging"
? "staging"
core/frontend/src/pages/queen-dm.tsx
+11 -2
View File
@@ -36,6 +36,7 @@ export default function QueenDM() {
const turnCounterRef = useRef(0);
const queenIterTextRef = useRef<Record<string, Record<number, string>>>({});
const [queenPhase, setQueenPhase] = useState<"planning" | "building" | "staging" | "running" | "independent">("independent");
// Switch queen session when queenId changes
useEffect(() => {
@@ -59,7 +60,7 @@ export default function QueenDM() {
(async () => {
try {
const result = await queensApi.getOrCreateSession(queenId);
const result = await queensApi.getOrCreateSession(queenId, undefined, "independent");
if (cancelled) return;
const sid = result.session_id;
@@ -196,6 +197,14 @@ export default function QueenDM() {
break;
}
case "queen_phase_changed": {
const rawPhase = event.data?.phase as string;
if (rawPhase === "independent" || rawPhase === "planning" || rawPhase === "building" || rawPhase === "staging" || rawPhase === "running") {
setQueenPhase(rawPhase);
}
break;
}
default:
break;
}
@@ -299,7 +308,7 @@ export default function QueenDM() {
isWaiting={isTyping && !isStreaming}
isBusy={isTyping}
disabled={loading || !queenReady}
queenPhase="planning"
queenPhase={queenPhase}
pendingQuestion={awaitingInput ? pendingQuestion : null}
pendingOptions={awaitingInput ? pendingOptions : null}
pendingQuestions={awaitingInput ? pendingQuestions : null}
linkedin_script.py
+16
View File
@@ -0,0 +1,16 @@
from playwright.sync_api import sync_playwright
def main():
with sync_playwright() as p:
browser = p.chromium.launch(headless=False)
page = browser.new_page()
page.goto("https://www.linkedin.com/login")
print("Please log in to LinkedIn in the opened browser window.")
input("Press Enter here when you have logged in...")
# Now search connections
print("Logged in. Ready to proceed.")
browser.close()
if __name__ == "__main__":
main()
scripts/llm_debug_log_visualizer.py
+149 -7
View File
@@ -878,13 +878,56 @@ def _sort_records(records: list[dict[str, Any]]) -> list[dict[str, Any]]:
)
def _load_session_data(
logs_dir: Path, session_id: str, limit_files: int
) -> list[dict[str, Any]] | None:
"""Load records for a specific session on demand."""
if not logs_dir.exists():
return None
files = sorted(
[path for path in logs_dir.iterdir() if path.is_file() and path.suffix == ".jsonl"],
key=lambda path: path.stat().st_mtime,
reverse=True,
)[:limit_files]
records: list[dict[str, Any]] = []
for path in files:
try:
with path.open(encoding="utf-8") as handle:
for line_number, raw_line in enumerate(handle, start=1):
line = raw_line.strip()
if not line:
continue
try:
payload = json.loads(line)
except json.JSONDecodeError:
payload = {
"timestamp": "",
"execution_id": "",
"assistant_text": "",
"_parse_error": f"{path.name}:{line_number}",
"_raw_line": line,
}
# Only include records for this session
if str(payload.get("execution_id") or "").strip() == session_id:
payload["_log_file"] = str(path)
records.append(payload)
except OSError:
continue
return _sort_records(records) if records else None
def _run_server(
html: str,
sessions: dict[str, list[dict[str, Any]]],
logs_dir: Path,
limit_files: int,
port: int,
no_open: bool,
) -> None:
html_bytes = html.encode("utf-8")
session_cache: dict[str, list[dict[str, Any]]] = {}
class Handler(http.server.BaseHTTPRequestHandler):
def do_GET(self) -> None:
@@ -892,11 +935,17 @@ def _run_server(
self._respond(200, "text/html; charset=utf-8", html_bytes)
elif self.path.startswith("/api/session/"):
sid = urllib.parse.unquote(self.path[len("/api/session/") :])
records = sessions.get(sid)
# Check cache first
if sid in session_cache:
records = session_cache[sid]
else:
records = _load_session_data(logs_dir, sid, limit_files)
if records is not None:
session_cache[sid] = records
if records is None:
self._respond(404, "application/json", b"[]")
else:
body = json.dumps(_sort_records(records), ensure_ascii=False).encode("utf-8")
body = json.dumps(records, ensure_ascii=False).encode("utf-8")
self._respond(200, "application/json", body)
else:
self.send_error(404)
@@ -927,13 +976,106 @@ def _run_server(
server.server_close()
def _discover_session_summaries(
logs_dir: Path, limit_files: int, include_tests: bool
) -> list[SessionSummary]:
"""Discover only session summaries without loading full record data."""
if not logs_dir.exists():
raise FileNotFoundError(f"log directory not found: {logs_dir}")
files = sorted(
[path for path in logs_dir.iterdir() if path.is_file() and path.suffix == ".jsonl"],
key=lambda path: path.stat().st_mtime,
reverse=True,
)[:limit_files]
# Collect minimal info per session: just first/last records and metadata
by_session: dict[str, list[dict[str, Any]]] = defaultdict(list)
for path in files:
try:
with path.open(encoding="utf-8") as handle:
for raw_line in handle:
line = raw_line.strip()
if not line:
continue
try:
payload = json.loads(line)
except json.JSONDecodeError:
continue
execution_id = str(payload.get("execution_id") or "").strip()
if execution_id:
# Store minimal data for summary generation
minimal = {
"timestamp": payload.get("timestamp", ""),
"iteration": payload.get("iteration", 0),
"stream_id": payload.get("stream_id", ""),
"node_id": payload.get("node_id", ""),
"token_counts": payload.get("token_counts", {}),
"_log_file": str(path),
}
by_session[execution_id].append(minimal)
except OSError:
continue
# Filter out test sessions if needed
if not include_tests:
by_session = {
eid: recs
for eid, recs in by_session.items()
if not _is_test_session(eid, recs)
}
summaries: list[SessionSummary] = []
for execution_id, session_records in by_session.items():
session_records.sort(
key=lambda record: (
str(record.get("timestamp", "")),
record.get("iteration", 0),
)
)
first = session_records[0]
last = session_records[-1]
summaries.append(
SessionSummary(
execution_id=execution_id,
log_file=str(first.get("_log_file", "")),
start_timestamp=str(first.get("timestamp", "")),
end_timestamp=str(last.get("timestamp", "")),
turn_count=len(session_records),
streams=sorted(
{str(r.get("stream_id", "")) for r in session_records if r.get("stream_id")}
),
nodes=sorted(
{str(r.get("node_id", "")) for r in session_records if r.get("node_id")}
),
models=sorted(
{
str(r.get("token_counts", {}).get("model", ""))
for r in session_records
if isinstance(r.get("token_counts"), dict)
and r.get("token_counts", {}).get("model")
}
),
)
)
summaries.sort(key=lambda summary: summary.start_timestamp, reverse=True)
return summaries
def main() -> int:
args = _parse_args()
records = _discover_records(args.logs_dir.expanduser(), args.limit_files)
summaries, sessions = _group_sessions(records, include_tests=args.include_tests)
logs_dir = args.logs_dir.expanduser()
# Only discover summaries, not full session data
summaries = _discover_session_summaries(
logs_dir, args.limit_files, args.include_tests
)
initial_session_id = args.session or (summaries[0].execution_id if summaries else "")
if initial_session_id and initial_session_id not in sessions:
if initial_session_id and not any(
s.execution_id == initial_session_id for s in summaries
):
print(f"session not found: {initial_session_id}")
return 1
@@ -945,7 +1087,7 @@ def main() -> int:
print(args.output)
return 0
_run_server(html_report, sessions, args.port, args.no_open)
_run_server(html_report, logs_dir, args.limit_files, args.port, args.no_open)
return 0