Orchestration Agent

Use this skill when the user wants to run, inspect, explain, or troubleshoot a Workflow Control workflow in this repository.

Workflow Control is a control plane. It creates run structure and tracks output files; it does not replace Product/QC/Engineering/Security/DevOps/Release skills.

Standard Run Layout

.agents/runs/<run-id>/
├── README.md
├── run.json
└── outputs/
    ├── 01-product-analysis.md
    ├── 02-design-handoff.md
    ├── 03-qc-test-plan.md
    ├── 04-engineering-handoff.md
    ├── 05-security-review.md
    ├── 06-devops-report.md
    └── 07-release-readiness.md

All agent outputs must be written under outputs/. Do not put step output files directly at the run root.

Workflow

1. Validate Workflow Control:

   python3 .agents/orchestration/scripts/workflow_runner.py validate

2. Pick the lightest workflow that fits the request:
   • technical_investigation: root cause only, no behavior change.
   • fast_path: docs, typo, local config, small refactor without behavior change.
   • full_sdlc: BRD/business/API/auth/data/medical/payment/production behavior.
   • adlc_prototype: AI/Agent/LLM/RAG/tool-calling feature before production.
3. Create run:

   python3 .agents/orchestration/scripts/workflow_runner.py init-run \
     --workflow full_sdlc \
     --title "<short title>" \
     --request "<original request>" \
     --project-context ".agents/project-contexts/<service-or-app>.context.md" \
     --source-type manual \
     --source-id "<stable id>"

4. For each step, run the owner skill and write its artifact to the path in run.json.
5. Check completion:
   • A step is done only when its expected output file exists.
   • A workflow is fully run only when all required step outputs exist or skipped optional steps have an explicit reason.

Reporting

Always report:

• Run ID and run folder.
• Output folder path.
• Completed steps versus total required steps.
• Missing output files.
• Whether this is READY, RUNNING, BLOCKED, or NOT_FULLY_RUN.