Plan Reviewer

Plan Reviewer

Minimum required inputs from the plan (reviewer gate)

Before reviewing details, confirm the plan includes these minimum inputs. If any are missing, request edits to the plan rather than guessing.

• Acceptance Criteria: clear, user visible outcomes, written as observable conditions
• Validation Plan: at least one validation step per acceptance criterion
• Automated Test Intent: what will be covered by automated tests, and where they live
• Manual Validation Steps: concrete steps anyone on the team can follow
• Integration Touchpoints: systems, services, queues, APIs, data stores, or clients that will be affected

Review implementation plans, PR specifications, and feature documentation before coding begins.

This skill focuses on:

• Testable acceptance criteria
• Complete validation strategies (manual + automated)
• Integration awareness
• Clear open questions and risks

It is designed to produce actionable deltas, not a perfect plan.

When to Use This Skill

Use this skill when reviewing:

• Implementation plans and execution checklists
• PR specifications
• Feature documentation and PRDs (where acceptance criteria exist)

Do not use this skill for:

• Reviewing actual code (use a code reviewer)
• Performance optimization reviews
• Security audits (use a dedicated security review)
• Architecture and design decisions (use an architecture review)

Definition of Done (Stop Conditions)

A plan review is “done” when:

1. Every acceptance criterion is testable
   Each acceptance criterion maps to at least one validation step.

2. Validation is complete enough to start coding
   Manual validation covers key scenarios, and automated tests are specified or explicitly deferred.

3. Integration points are explicit and validated
   Anything the change touches has a corresponding verification step.

4. Open questions are documented
   Any ambiguity is called out as a decision, not silently assumed.

5. Top risks are listed with mitigations
   At least the top 3 risks are identified and tied to validation.

Once these five conditions are met, stop expanding.

Core Review Process (Always Run)

Step 1: Acceptance Criteria Testability

For each acceptance criterion, check:

• ✅ Is it specific and measurable?
• ✅ Can it be validated manually?
• ✅ Can it be validated automatically (unit, integration, UI)?
• ❓ If not testable, what decision or rewrite is required?

Common gaps to flag

• Vague language: “works”, “handles gracefully”, “clearly”, “fast”
• Unbounded scope: “supports all formats”, “all edge cases”
• Missing constraints: timeouts, retries, limits, permissions

Rule: Do not invent new product requirements.
Only make existing requirements testable, or flag them as ambiguous.

Step 2: Manual Validation Completeness

Manual validation should cover:

• Happy path
• At least one realistic failure path
• At least one integration touchpoint
• Expected results, not just “verify it works”

Manual validation steps should include:

• Preconditions (environment, flags, account state, data)
• Step by step actions
• Expected outcome
• Observability (logs, metrics, UI state, API responses)

Step 3: Automated Test Expectations (Actionable)

Convert vague testing statements into explicit expectations:

• Where tests live (file or suite)
• What is covered (scenario level)
• What is mocked vs real (dependency strategy)

Examples:

• “Unit tests cover X” → list concrete test cases
• “Integration tests exist” → describe workflow boundaries (POST → GET → UPDATE)
• “Error cases are tested” → specify which errors and expected responses

Naming guidance Use descriptive names, consistent with existing patterns:

• test_create_resource_when_valid_returns_201()
• test_create_resource_when_missing_field_returns_400()
• test_service_retries_on_timeout_then_fails()

Step 4: Integration Points and Required Validations

List the systems touched by this change, then for each one specify:

• What could break
• How we validate it

Examples of integration points:

• API endpoints and clients
• Persistent storage
• Configuration and flags
• UI surfaces
• External dependencies (third party APIs, framewors)

Outcome: a short “integration validation checklist” tied to the plan.

Optional Modules (Only If Relevant)

Use these checklists when the plan includes the relevant domain.

Module: Configuration and Precedence

Verify:

• Which sources exist (env, file json, DB, secret manager, etc.)
• Precedence rules are explicitly defined
• Behavior when a value is missing, invalid, or changed

Add validations:

• precedence tests
• unset / rollback behavior
• failure handling

Module: Backward Compatibility and Safe Defaults

Verify:

• old clients or old data still function
• do we need to be backwards compatible, or is it OK to make breaking changes (KISS, SRP, DRY, YAGNI - prefer to keep logic simple)
• missing fields have defaults
• schema changes are safe or migrated

Add validations:

• old payloads accepted or rejected explicitly
• old reads remain stable
• compatibility expectations are documented

Module: External Dependency Failure

Verify:

• timeouts and retries are specified
• errors are surfaced cleanly
• user experience is defined during partial outages

Add validations:

• simulate dependency outage
• verify error messages
• verify no infinite retry loops

Module: UI Workflow (If Applicable)

Verify:

• loading state is defined
• error state is defined
• empty state is defined
• state consistency matches API behavior

Add validations:

• slow API simulation
• error responses
• empty dataset behavior

Module: Security and Secrets Hygiene

Verify:

• no secrets are logged
• no secrets are returned in API responses
• sensitive values are masked consistently

Add validations:

• redaction tests
• response content checks
• log output checks

Risk Scoring (PM + Principal Engineer Friendly)

For each major risk, record:

• Impact: Low / Medium / High
• Likelihood: Low / Medium / High
• Mitigation: what test or validation catches it

Example:

• Risk: “Old clients break due to response change”
  Impact: High
  Likelihood: Medium
  Mitigation: integration test with old payload, contract tests

Keep this list short: top 3 to 5 risks.

Default Review Output (Short)

Use this structure by default:

# Plan Review: [Feature Name]

## 1. Readiness Verdict
- Verdict: ✅ Ready to implement / 🟡 Ready after fixes / 🔴 Not ready
- Why: [1 to 3 bullets]

## 2. Open Questions To Resolve Before Coding (Decisions needed)
- [ ] Item 1
- [ ] Item 2

## 3. Open Requirements
### Must resolve before coding (blocking)
- Items where the acceptance criterion itself is missing, ambiguous, or untestable

### Resolve during implementation (non-blocking)
- Items where the WHAT is clear from the acceptance criteria but the exact HOW is an implementation detail
- Example: "a test seam is needed for year injection" is non-blocking if the AC already says "rejects non-2026 year before writes"
- Example: "which exact mock library to use" is non-blocking

## 4. Validation Plan Summary
### Manual
- Scenario: ...
   - [ ] Item 1
   - [ ] Item 2
   ...
- Scenario: ...
   ...

### Automated
- Unit: ...
   - [ ] Item 1
   - [ ] Item 2
   ...
- Integration: ...
   - [ ] Item 1
   - [ ] Item 2
   ...
- UI (if applicable): ...
   - [ ] Item 1
   - [ ] Item 2
   ...

## 5. Top Risks and Mitigations
- Risk: ...
  - Impact: ...
  - Likelihood: ...
  - Mitigation: ...

Expanded Output (Only When Requested)

If the reviewer requests a full deep dive, expand into:

• Acceptance criteria gaps
• Manual validation gaps
• Automated testing specifications
• Integration validation checklist
• Compatibility and migration notes
• Prioritized recommendations
• Unresolved question and for each one, iterate with reviewer/human until decision is made, which is then documented in the plan.

Principles

1. Testability first: Every acceptance criterion must be testable
2. Specificity over generality: “Test X” is not enough, specify what and how
3. Stop conditions matter: Finish when the plan is shippable, not perfect
4. Integration awareness: Verify what could break outside the happy path
5. Do not invent requirements: Flag ambiguity, do not hallucinate scope
6. Prefer actionable deltas: Smallest set of changes that unblock coding

Automated Test Expectations

Requirement: For each Acceptance Criterion, provide at least one explicit example test case when feasible. If a criterion cannot be reasonably automated, mark it as Deferred and include the reason and what manual validation will cover.

When a test approach requires codebase investigation to determine the mechanism (e.g., how to inject a test seam, which mock strategy to use), classify it as "resolve during implementation" rather than blocking — as long as the plan states the acceptance criterion the test must satisfy.