Review Plan

Critically review implementation plans using Codex CLI with high-reasoning capabilities.

Purpose

Validate implementation plans to ensure they:

• Make logical sense and are achievable
• Follow project standards and conventions
• Have proper structure and completeness
• Won't cause issues during implementation

When to Use

• Before executing a plan with implement-plan
• After creating a plan with create-plan
• When reviewing someone else's plan
• To validate plan updates or changes

CRITICAL: Third-Party Review Required

This skill MUST delegate to Codex CLI. The review cannot be performed by the agent creating the plan - an independent third-party model must validate the work.

Why third-party review matters:

• Catches blind spots the author missed
• Validates assumptions against actual codebase state
• Provides independent verification of feasibility
• Reduces confirmation bias in self-review

If Codex CLI is not available, inform the user and do not proceed with a self-review.

Codex CLI Invocation

Use the dedicated codex review command for non-interactive review:

# Pass instructions directly as argument
codex review "Review the plan at docs/plans/0042_feature.md for feasibility and completeness"

# Or read instructions from stdin
cat <<'PROMPT' | codex review -
Review the implementation plan at docs/plans/0042_feature.md

Check for: logical coherence, completeness, standards compliance, feasibility.
Provide PASS/FAIL/NEEDS_REVISION verdict.
PROMPT

Key options:

• Use [PROMPT] argument for simple instructions
• Use - to read longer instructions from stdin
• Use heredoc with 'PROMPT' (quoted) to prevent variable expansion

Note: When reviewing plans (not code changes), use the [PROMPT] form since --base, --commit, and --uncommitted flags are for reviewing code diffs.

Workflow

Step 1: Locate the Plan

If plan path provided: Use the specified path directly.

If no path provided:

1. Check current conversation context for a plan being discussed
2. Look in docs/plans/ for the most recently modified plan
3. Ask user which plan to review

Step 2: Gather Context

Before invoking Codex, gather relevant context:

1. Read the plan file completely
2. Identify referenced files mentioned in the plan
3. Load relevant standards:
   • Project instructions: CLAUDE.md or AGENTS.md
   • Applicable standards from content/standards/ based on tech stack
4. Explore codebase as needed to validate plan assumptions

Step 3: Invoke Codex Review

MANDATORY: You must invoke Codex CLI. Do not perform the review yourself.

First, verify Codex is available:

which codex || echo "Codex CLI not installed"

If not available, stop and inform the user they need to install Codex CLI.

Build the review prompt and invoke Codex using stdin:

cat <<'PROMPT' | codex review -
Review the implementation plan at {plan-path}

Check against these criteria:
1. Logical Coherence - Do phases flow logically? Dependencies correct?
2. Completeness - All steps included? Testing strategy adequate?
3. Standards Compliance - Follows project conventions?
4. Feasibility - File paths correct? Referenced files exist?
5. Risk Assessment - What could go wrong?

Provide:
1. PASS/FAIL/NEEDS_REVISION verdict
2. Summary of findings
3. Specific issues with severity (HIGH/MEDIUM/LOW)
4. Recommended changes
PROMPT

Important: Let Codex read the plan file itself - it has filesystem access. This allows it to:

• Verify file paths mentioned in the plan actually exist
• Check referenced modules and their exports
• Validate assumptions about codebase state

Do NOT paste plan content into the prompt - this wastes tokens and prevents Codex from exploring the codebase.

Step 4: Compile Review Report

After Codex returns, format the results:

## Plan Review: {plan-name}

**Verdict:** PASS | FAIL | NEEDS_REVISION
**Reviewed by:** Codex (gpt-5.1-codex-max)
**Date:** {timestamp}

### Summary
{Brief overview of findings}

### Issues Found

| Severity | Issue | Location | Recommendation |
|----------|-------|----------|----------------|
| HIGH | {issue} | {section} | {fix} |
| MEDIUM | {issue} | {section} | {fix} |
| LOW | {issue} | {section} | {fix} |

### Standards Compliance
- [x] Follows project conventions
- [ ] Testing strategy adequate
- [x] Architecture aligned

### Questions for Author
1. {clarifying question}

### Recommendation
{Next steps - proceed, revise, or reject}

Review Criteria Reference

See references/review-criteria.md for the complete checklist.

Standards Sources

The review checks against:

1. Project Instructions

   • CLAUDE.md or AGENTS.md in project root
   • .claude/ configuration

2. Content Standards (based on tech stack)

   • content/standards/backend/java/ - Java backend standards
   • content/standards/frontend/react/ - React frontend standards

3. Plan Format Requirements

   • Token limit: < 5000 tokens
   • Required sections: Testing Strategy, Dependency Graph
   • Proper numbering: {NNNN}_{name}.md

Examples

Example: Review a specific plan

User: Review the plan at docs/plans/0042_user-auth.md

Claude: I'll review this plan using Codex CLI for independent validation.

[Runs: which codex]
/opt/homebrew/bin/codex

[Runs: cat <<'PROMPT' | codex review -
Review the implementation plan at docs/plans/0042_user-auth.md
...
PROMPT]

Codex returned:

Verdict: NEEDS_REVISION

Summary: The plan is well-structured but missing integration tests
and has an incorrect dependency ordering in Phase 3.

Issues:
- HIGH — Missing integration tests (Testing Strategy section)
- MEDIUM — Task 3.2 depends on 3.4 but is ordered before it (Phase 3)

Recommended changes:
- Add API integration tests to Testing Strategy
- Reorder Phase 3 tasks so 3.4 comes before 3.2

---

Based on Codex's review, you should address the HIGH severity issue
before proceeding with implementation.

Example: Codex not available

User: Review the plan

Claude: I need to invoke Codex CLI for third-party review.

[Runs: which codex]
codex not found

I cannot proceed with the review. This skill requires Codex CLI
for independent third-party validation. Please install it:

  npm install -g @openai/codex

Then run /review-plan again.

Output

The skill produces:

• Formatted review report in conversation
• Verdict: PASS, FAIL, or NEEDS_REVISION
• Specific actionable feedback

Related Skills

• create-plan - Creates plans that this skill reviews
• implement-plan - Executes plans after review passes
• skill-validator - Similar validation pattern for skills