plan-validate

Plan Validate

Deterministic plan.json validation that replaces the plan-validator Sonnet agent for all structural checks. The semantic consistency check (storage/query alignment, task/architecture contradictions) still requires the LLM agent and should be run separately when needed.

When to Use

Activate this skill instead of launching @code:plan-validator at every plan validation site. The orchestrator should only launch the full plan-validator agent for semantic-only checks after plan creation or modification phases.

Usage

Run the validation script:

python3 ${CLAUDE_SKILL_DIR}/scripts/validate_plan.py <WORKDIR> --auto-sync

The --auto-sync flag detects questions answered in the markdown content that are still in the openQuestions JSON array, extracts the answer text, and moves them to answeredQuestions before validation — writing the updated plan.json back to disk.

Three answer formats are supported:

1. Inline with prefix — **Answer: text**, *Answer: text*, or plain Answer: text on the question line.
2. A-### keyed answer — a separate A-001: answer text line corresponding to Q-001.
3. Inline comment — extra text appended after the known question text on the question line, without any Answer: prefix. Metadata markers like (BLOCKING T-X.Y) and [Recommended: ...] are stripped.

The question checkbox does not need to be checked for any format. When a question is migrated from an unchecked [ ] line, the script checks it automatically.

If none of these yield answer text, falls back to the recommendedAnswer field from the JSON entry. Questions with no extractable answer are left in openQuestions.

Always pass --auto-sync so that users can answer questions by editing the markdown directly.

Interpreting Output

The script prints JSON to stdout matching the exact plan-validator output format.

Plan Valid (PLAN_VALID)

{
  "status": "VALID",
  "issues": [],
  "has_unanswered_questions": false,
  "unanswered_questions": [],
  "has_answered_questions": false,
  "answered_questions": [],
  "has_addressed_gaps": false,
  "addressed_gaps": [],
  "pending_tasks": [{"id": "T-1.1", "description": "...", "acceptanceCriteria": ["AC-001"]}],
  "completed_tasks": [],
  "manual_tasks": [],
  "decision_table_path": ".closedloop-ai/decision-tables/pln-001.md",
  "decision_table_status": "pending"
}