Plan Enforcer Review

Plan Enforcer Review

Use this before execution when the user wants a plan quality check.

Public front-end flow is discuss -> draft -> review.

Goal

Review a plan file or plan text and answer one question:

Is this plan safe enough to enforce as-is?

Plan Enforcer should not blindly enforce weak plans. Review first when quality is uncertain.

What To Check

Look for:

• vague tasks ("cleanup", "improve", "refactor" without a concrete target)
• missing verification steps
• tasks that are too broad or hide multiple changes
• sequencing risks (ship/deploy before verification)
• migration/integration work without explicit dependencies
• missing guardrails such as assumptions, constraints, or out-of-scope notes
• docs/examples that look like plan tasks but are not meant to execute
• drift from the discuss packet at .plan-enforcer/discuss.md

(or the legacy .plan-enforcer/combobulate.md fallback) when an intent packet exists:

• dropped non-negotiables
• missing proof requirements
• forbidden narrowings adopted by the plan
• goal text that no longer preserves the normalized goal

Output Shape

Start with a verdict:

• pass -- safe to execute
• weak -- executable, but needs edits first
• unsafe -- do not execute yet

Then list findings in severity order with concrete fixes.

Preferred structure:

Verdict: weak

Findings:
- T2 is vague: "Improve backend" -> name the exact subsystem and intended behavior
- T3 has no verification step -> add a concrete check, test, or manual proof
- Deploy appears before verification -> move release after the proving step

Auto-repair suggestions:
- Rewrite T2 as: "Update the Express auth middleware to reject expired sessions."
- Add under T3: "- [ ] Verify `npm test -- auth` passes"

Suggested repair block:

Behavior

• At stage start, set the statusline stage to 3-REVIEW with:

node "$HOME/.claude/skills/plan-enforcer/src/statusline-stage-cli.js" review --label 3-REVIEW

• If the user provides a path, read that plan.
• If the user provides inline plan text, review it directly.
• If the installed CLI is available, prefer running node ~/.claude/skills/plan-enforcer/src/review-cli.js so the verdict, findings, and repair block match the shared formatter exactly.
• When reviewing a file path, also look for the discuss packet at

.plan-enforcer/discuss.md near the repo root, falling back to the legacy .plan-enforcer/combobulate.md path when needed, and treat packet-to-plan drift as a first-class review finding.

• If the plan is weak or unsafe, propose the minimum repair needed.
• Prefer specific replacement text over abstract advice when you can.
• If one or two tasks are broken, rewrite just those tasks.
• If the whole plan is weak, provide a short repair block the user can paste back into the plan.
• Always include a Suggested repair block: section when the verdict is weak or unsafe.
• Do not execute the plan while reviewing it unless the user explicitly asks for both review and execution.

Positioning

This is not a heavyweight planning framework. It is a guardrail:

• discuss clarifies ambiguous asks first
• draft writes the plan
• review checks the plan
• enforcer keeps execution honest

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jccidc
• Source: jccidc/.plan-enforcer
• License: MIT
• Homepage: https://www.jccidc.com/

Install and usage instructions live in the source repository linked above.