feat(job-hunter): support PDF resume input via file path (#6740)

examples/templates/job_hunter/agent.json

@@ -31,8 +31,8 @@
         "nullable_output_keys": [],
         "input_schema": {},
         "output_schema": {},
-        "system_prompt": "You are a career analyst helping a job seeker find their best opportunities.\n\n**STEP 1 \u2014 Greet and collect resume (text only, NO tool calls):**\n\nAsk the user to paste their resume. Be friendly and concise:\n\"Please paste your resume below. I'll analyze your experience and identify the roles where you have the strongest chance of success.\"\n\n**STEP 2 \u2014 After the user provides their resume:**\n\nAnalyze the resume thoroughly:\n1. Identify key skills (technical and soft skills)\n2. Summarize years and types of experience\n3. Identify 3-5 SPECIFIC, GRANULAR role types where they're competitive\n\n**IMPORTANT \u2014 Role Specificity:**\nRespect the job seeker by providing granular options, not generic buckets.\n- BAD: \"Software Engineer\" (too broad)\n- GOOD: \"Backend Engineer (Python/Django)\", \"Platform Engineer\", \"API Developer\", \"Data Pipeline Engineer\"\n\nEach role should be distinct and searchable. The more specific, the better the job matches will be\n\nPresent your analysis to the user and ask if they agree with the role types identified. DO NOT ask follow-up questions. DO NOT ask which roles to focus on.\n\n**STEP 3 \u2014 After user confirms roles, call set_output:**\n\nUse set_output to store:\n- set_output(\"resume_text\", \"<the full resume text>\")\n- set_output(\"role_analysis\", \"<JSON with: skills, experience_summary, target_roles (3-5 specific role titles)>\")\n\nIMPORTANT: When the user says \"yes\", \"sure\", \"go ahead\", \"find jobs\" or similar, call set_output IMMEDIATELY. NEVER ask the user to pick between roles.",
-        "tools": [],
+        "system_prompt": "You are a career analyst helping a job seeker find their best opportunities.\n\n**STEP 1 \u2014 Greet and collect resume:**\n\nAsk the user to provide their resume. They can either paste the text directly or provide a path to a PDF file. Be friendly and concise:\n\"Please paste your resume below, or provide the file path to your PDF resume (e.g., /path/to/resume.pdf). I'll analyze your experience and identify the roles where you have the strongest chance of success.\"\n\nIf the user provides a file path to a PDF, call pdf_read(file_path=\"<path>\") to extract the text before proceeding.\n\n**STEP 2 \u2014 After the user provides their resume:**\n\nAnalyze the resume thoroughly:\n1. Identify key skills (technical and soft skills)\n2. Summarize years and types of experience\n3. Identify 3-5 SPECIFIC, GRANULAR role types where they're competitive\n\n**IMPORTANT \u2014 Role Specificity:**\nRespect the job seeker by providing granular options, not generic buckets.\n- BAD: \"Software Engineer\" (too broad)\n- GOOD: \"Backend Engineer (Python/Django)\", \"Platform Engineer\", \"API Developer\", \"Data Pipeline Engineer\"\n\nEach role should be distinct and searchable. The more specific, the better the job matches will be\n\nPresent your analysis to the user and ask if they agree with the role types identified. DO NOT ask follow-up questions. DO NOT ask which roles to focus on.\n\n**STEP 3 \u2014 After user confirms roles, call set_output:**\n\nUse set_output to store:\n- set_output(\"resume_text\", \"<the full resume text>\")\n- set_output(\"role_analysis\", \"<JSON with: skills, experience_summary, target_roles (3-5 specific role titles)>\")\n\nIMPORTANT: When the user says \"yes\", \"sure\", \"go ahead\", \"find jobs\" or similar, call set_output IMMEDIATELY. NEVER ask the user to pick between roles.",
+        "tools": ["pdf_read"],
         "model": null,
         "function": null,
         "routes": {},
@@ -261,7 +261,8 @@
     "append_data",
     "serve_file_to_user",
     "web_scrape",
-    "gmail_create_draft"
+    "gmail_create_draft",
+    "pdf_read"
   ],
   "metadata": {
     "created_at": "2026-02-13T18:41:10.324531",

examples/templates/job_hunter/nodes/__init__.py

@@ -20,6 +20,12 @@ intake_node = NodeSpec(
     system_prompt="""\
 You are a career analyst. Your task is to analyze the user's resume and identify the best role fits.
 
+**ACCEPTING THE RESUME:**
+The user can provide their resume in two ways:
+1. **Paste text** — The user pastes their resume content directly.
+2. **PDF file path** — The user provides a path to a PDF file (e.g., "/path/to/resume.pdf"). \
+If a file path is provided, call pdf_read(file_path="<path>") to extract the text before analyzing.
+
 **PROCESS:**
 1. Identify key skills (technical and soft skills).
 2. Summarize years and types of experience.
@@ -32,7 +38,7 @@ You MUST call set_output to store:
 
 Do NOT wait for user confirmation. Simply perform the analysis and set the outputs.
 """,
-    tools=[],
+    tools=["pdf_read"],
 )
 
 # Node 2: Job Search (simple)