factory-root-planner

name	factory-root-planner
description	Coordinate a Factory run from raw brief through Stage I2. Use when Codex is asked to initialize a Factory run, run Stage A, create or repair run-root evidence, enforce Factory read order, choose PLANNING_ONLY versus EXECUTION_ENABLED posture, coordinate the full stage flow, or prepare a pack for human review without executing implementation.

Factory Root Planner

Workflow

1. Read AGENTS.md, docs/Factory/ORCHESTRATION.md, docs/Factory/Spec/STAGE_CONTRACTS.md, and docs/Factory/SCRATCHPAD.md ## Active Pitfalls (Mandatory).
2. Run bash scripts/knowledge_lint.sh before Stage A and persist output as docs/Factory/runs/<RUN_ID>/KNOWLEDGE_LINT.txt.
3. Create the run root and required files: raw_brief.md, EXECUTION_MODE.txt, CONTEXT_RECALL_REPORT.md, and later SPRINT_ID.txt.
4. Build recall evidence with ./scripts/factoryctl context-index and ./scripts/factoryctl context-report --profile stage-a.
5. Keep the run PLANNING_ONLY unless the raw brief explicitly authorizes EXECUTION_ENABLED.
6. Coordinate stages in order: A -> B -> C -> D -> E -> F -> G -> H -> I -> J -> I2.
7. After each stage handoff, run ./scripts/factoryctl stage-lint --run <RUN_ID> --stage <STAGE>.
8. After I2, run ./scripts/factoryctl pack-lint --run <RUN_ID> before human review.

Guardrails

• Do not execute sprint implementation during Factory planning.
• Do not expand scope silently; record proposed expansion as BLOCKING unless approved.
• Treat recall artifacts as evidence aids, not authority.
• If Coverage Verdict: WEAK, halt and repair recall before drafting intent.
• If any validator fails, fix the artifact or re-run the affected stage before advancing.

Outputs

Return:

• run id
• execution mode
• evidence paths created
• current stage status
• validator results
• blockers requiring human decision