Plan Review

Use this skill to validate implementation plans across completeness, quality, feasibility, and scope dimensions.

When to Use

After plan is written and user selects "A) review the plan" option.

Phase 1: Initial Assessment

Run automatic checks across 4 dimensions using simple validation logic (no subagents yet):

Completeness Check

Scan plan for:

• ✅ All phases have success criteria section
• ✅ Commands for verification present (make test-, pytest, etc.)
• ✅ Rollback/migration strategy mentioned
• ✅ Edge cases section or error handling
• ✅ Testing strategy defined

Scoring:

• PASS: All criteria present
• WARN: 1-2 criteria missing
• FAIL: 3+ criteria missing

Quality Check

Scan plan for:

• ✅ File paths with line numbers: file.py:123
• ✅ Specific function/class names
• ✅ Code examples are complete (not pseudocode)
• ✅ Success criteria are measurable
• ❌ Vague language: "properly", "correctly", "handle", "add validation" without specifics

Scoring:

• PASS: File paths present, code complete, criteria measurable, no vague language
• WARN: Some file paths missing or minor vagueness
• FAIL: No file paths, pseudocode only, vague criteria

Feasibility Check

Basic checks (detailed check needs subagent):

• ✅ References to existing files/functions seem reasonable
• ✅ No obvious impossibilities
• ✅ Technology choices are compatible
• ✅ Libraries mentioned are standard/available

Scoring:

• PASS: Seems feasible on surface
• WARN: Some questionable assumptions
• FAIL: Obvious blockers or impossibilities

Scope Creep Check

Requires research.md memory or brainstorm context:

• ✅ "What We're NOT Doing" section exists
• ✅ Features align with original brainstorm
• ❌ New features added without justification
• ❌ Gold-plating or over-engineering patterns

Scoring:

• PASS: Scope aligned with original decisions
• WARN: Minor scope expansion, can justify
• FAIL: Significant scope creep or gold-plating

Phase 2: Escalation (If Needed)

If any dimension scores FAIL, spawn specialized validators:

const failedDimensions = {
  completeness: score === 'FAIL',
  quality: score === 'FAIL',
  feasibility: score === 'FAIL',
  scope: score === 'FAIL'
}

// Spawn validators in parallel for failed dimensions
const validations = await Promise.all([
  ...(failedDimensions.completeness ? [Task({
    subagent_type: "completeness-checker",
    description: "Validate plan completeness",
    prompt: `
      Analyze this implementation plan for completeness.

      Plan file: ${planPath}

      Check for:
      - Success criteria (automated + manual)
      - Dependencies between phases
      - Rollback/migration strategy
      - Edge cases and error handling
      - Testing strategy

      Report issues and recommendations.
    `
  })] : []),

  ...(failedDimensions.feasibility ? [Task({
    subagent_type: "feasibility-analyzer",
    description: "Verify plan feasibility",
    prompt: `
      Verify this implementation plan is feasible.

      Plan file: ${planPath}

      Use Serena MCP to check:
      - All referenced files/functions exist
      - Libraries are in dependencies
      - Integration points match reality
      - No technical blockers

      Report what doesn't exist or doesn't match assumptions.
    `
  })] : []),

  ...(failedDimensions.scope ? [Task({
    subagent_type: "scope-creep-detector",
    description: "Check scope alignment",
    prompt: `
      Compare plan against original brainstorm for scope creep.

      Plan file: ${planPath}
      Research/brainstorm: ${researchMemoryPath}

      Check for:
      - Features not in original scope
      - Gold-plating or over-engineering
      - "While we're at it" additions
      - Violations of "What We're NOT Doing"

      Report scope expansions and recommend removals.
    `
  })] : []),

  ...(failedDimensions.quality ? [Task({
    subagent_type: "quality-validator",
    description: "Validate plan quality",
    prompt: `
      Check this implementation plan for quality issues.

      Plan file: ${planPath}

      Check for:
      - Vague language vs. specific actions
      - Missing file:line references
      - Untestable success criteria
      - Incomplete code examples

      Report specific quality issues and improvements.
    `
  })] : [])
])

Phase 3: Interactive Refinement

Present findings conversationally (like brainstorming skill):

I've reviewed the plan. Here's what I found:

**Completeness: ${score}**
${if issues:}
- ${issue-1}
- ${issue-2}

**Quality: ${score}**
${if issues:}
- ${issue-1}
- ${issue-2}

**Feasibility: ${score}**
${if issues:}
- ${issue-1}
- ${issue-2}

**Scope: ${score}**
${if issues:}
- ${issue-1}
- ${issue-2}

${if any FAIL:}
Let's address these issues. Starting with ${most-critical-dimension}:

Q1: ${specific-question}
   A) ${option-1}
   B) ${option-2}
   C) ${option-3}

Question Flow

Ask one question at a time, wait for answer, then next question.

For each issue:

1. Explain the problem clearly
2. Offer 2-4 concrete options
3. Allow "other" for custom response
4. Apply user's decision immediately
5. Update plan if changes agreed
6. Move to next issue

Refinement Loop

After addressing all issues:

1. Update plan file with agreed changes
2. Re-run Phase 1 assessment
3. If still FAIL, spawn relevant validators again
4. Continue until all dimensions PASS or user approves WARN

Approval

When all dimensions PASS or user accepts WARN:

Plan review complete! ✅

**Final Scores:**
- Completeness: PASS
- Quality: PASS
- Feasibility: PASS
- Scope: PASS

The plan is ready for execution.

If user approved with WARN:

Plan review complete with warnings.

**Final Scores:**
- Completeness: PASS
- Quality: WARN (minor file path gaps)
- Feasibility: PASS
- Scope: PASS

Warnings acknowledged. Proceeding with execution.

Plan Updates

When making changes based on review:

1. Read current plan file
2. Apply specific edits using Edit tool
3. Preserve structure and formatting
4. Update only sections with issues
5. Verify changes with re-read

Don't rewrite entire plan - use targeted edits.

Exit Conditions

Success: All PASS or user approves WARN Abort: User requests to skip review and execute anyway Restart: User requests major redesign (back to brainstorm)

Example Session

I've reviewed the plan. Here's what I found:

**Completeness: ⚠️ WARN**
- Phase 2 lacks automated success criteria

**Quality: ✅ PASS**

**Feasibility: ❌ FAIL**
- Plan assumes `src/auth/handler.py` exists - NOT FOUND

**Scope: ❌ FAIL**
- Plan includes "admin dashboard" but brainstorm only mentioned "user dashboard"

Let's address these issues. Starting with scope:

Q1: The plan includes an admin dashboard, but your brainstorm only mentioned a user dashboard. Should we:
   A) Remove admin dashboard (stick to original scope)
   B) Keep it (intentional expansion)
   C) Split admin dashboard into separate plan

User: A

Got it. I'll remove the admin dashboard from the plan.

[Updates plan, removes admin dashboard task]

Next, feasibility:

Q2: The plan references `src/auth/handler.py` but this file doesn't exist. I found `src/security/auth.py:45` with similar functionality. Should we:
   A) Create new `src/auth/handler.py` as planned
   B) Use existing `src/security/auth.py` instead
   C) Rename existing file to match plan

User: B

Makes sense. I'll update the plan to use `src/security/auth.py`.

[Updates plan with correct file paths]

Finally, completeness:

Q3: Phase 2 doesn't have automated success criteria. Should we add:
   A) `make test-auth` command to verify
   B) Specific pytest command
   C) Manual testing steps only

User: A

Perfect. I'll add `make test-auth` verification to Phase 2.

[Updates plan with verification command]

Re-checking plan with fixes...

**Final Scores:**
- Completeness: ✅ PASS
- Quality: ✅ PASS
- Feasibility: ✅ PASS
- Scope: ✅ PASS

Plan review complete! The plan is ready for execution.