plan-work-completion-signoff

Plan Work Completion Signoff

Verify that implemented plan docs match their actual code. For each plan doc, an agent compares every specified feature, API, type, data structure, and test category against the real implementation. Complete docs get a signoff section appended; docs with gaps generate a report so the orchestrator can create follow-up beads.

This skill is a structured decision framework for the scheduler/concierge agent — it orchestrates multi-agent verification work. Individual agents do the actual comparison and signoff.

Inputs

• $0 (optional): Plans directory (default: docs/plans/)
• $1 (optional): Code base path (default: repo root)

Phase 1: Discover Implemented Plans

1. Read the plan index (docs/plans/00-plan-index.md or equivalent)
2. Identify which plan docs have been implemented — look for:
   • Closed implementation beads/epics referencing those plans
   • Existing code packages that correspond to plan components
   • Plan doc status markers (e.g., "Implementation complete" in the index)
3. Build a list of plan doc pairs to verify: each plan doc + its companion test harness doc (if exists)
4. Exclude docs that already have a ## Completion Signoff section (already verified in a prior pass)

Decision: Communicate the discovered doc list to the user or concierge for confirmation before proceeding. If the list looks wrong (too many or too few docs), clarify before creating beads.

Phase 2: Create Beads and Assign Agents

1. Create an epic bead: bd create "Plan completion signoff" --type epic --labels project={project}
2. Group plan docs into tasks — aim for 2-4 docs per task, grouped by component area:
   • Group a plan doc with its companion test harness doc in the same task
   • Related components can share a task (e.g., a storage layer plan + its test harness plan)
   • Don't make tasks too small (one doc each) or too large (8+ docs)
3. Create task beads under the epic, one per group
4. Assign tasks to available agents. Prefer agents who:
   • Wrote the implementation (they know the code best)
   • Reviewed the implementation (they know the gaps)
   • If original agents are unavailable, any agent can do it — the plan docs and code are self-documenting