| name | resume-orchestration |
| description | Execution mode - Workflow for generating tailored resumes and cover letters based on a Job Description (JD). Use this when the user asks to create a resume for a specific job or provides a position description. |
CRITICAL INSTRUCTION: There are model and runtime-specific variances for agents running in this environment. If you are a Gemini Agent in Gemini CLI, you will use Gemini_Tasks. If you are an OpenAI Agent running under Codex, you will use Codex_Tasks.
Resume Orchestration Workflow
This workflow automates the process of creating a tailored JSON Resume, HTML/PDF exports, and a Cover Letter.
WARNING: RESOURCE INTENSIVE OPERATION
- This is a time and token expensive operation.
- It can take between 15 to 45 minutes for the thorough process to be completed.
- It involves multiple operations with 8 separate Gemini-based AI Agents.
- INSTRUCTION FOR AI: You MUST explicitly warn the user about this 15-45 minute execution time before you start the orchestrator or immediately upon starting it, so they know what to expect.
Prerequisites
- Job Description (JD) saved as a
.txt file (e.g., in cv-data/resumes/).
- Python virtual environment:
./venv/bin/python3.
Execution
Run the orchestrator using the following command. Do not background this process, use timeout, or attempt to tail logs. Simply execute the script and wait for it to complete.
./venv/bin/python3 AGENT_Tasks/Orchestrator/resume_orchestrator.py --jd <path_to_job_description.txt>
Optional Arguments:
--notes "Strategic guidance": Provide extra context to the builder (e.g., "Emphasize AI and Security"). These are notes provided by user to the resume builder. You are never to make decisions on what should be added. The Resume Builder has instructions which should be modified by the user and not by the AI.--sentinel-only: Only run the initial security/trap detection check.--resume-from-audit: Use this flag to resume the pipeline directly at Phase 3 (Audit Loop). This safely restarts a failed run by: 1) Loading a fresh context model first, 2) Skipping Phase 1 and 2, and 3) Reusing context and the cache of previously verified items while discarding any unaddressed in-progress exceptions. It is only safe to resume if the previous run completed Phase 2 and you have a valid JSON draft to audit. Do not resume if the run failed after the Forensic Audit but before the Fixer, as the active changes list would be lost.
Post-Execution
Once the orchestrator completes successfully:
- Stage all changes in the
cv-data/ repository directory.
- Stage all changes in the main
Cv repository.
- Commit and push the changes for both repositories. Use a descriptive commit message indicating the generation of the resume for the specific JD.
Workflow Phases
- Phase 0 Setup: Initializes master session, calculates word frequency, and loads CV data and JD. Includes ASCII Sanitization (via
unidecode) of all inputs to ensure strict ASCII compatibility.
- Phase 1 Sentinel: Performs a security threat check on the JD.
- Phase 2 Builder: Generates an initial JSON draft tailored to the JD.
- Phase 3 Validate Audit Review: Performs schema validation, forensic audit, persona review, and ASCII Validation. Any detected Unicode triggers a fixer loop or manual audit requirement.
- Phase 4 Cover Letter: Generates a matching cover letter.
- Phase 5 Export Artifacts: Renders the resume to HTML and PDF (using 'stackoverflow' theme) and generates an appendix PDF.
Output Handling
