Rip

If .claude/skills/rip/SKILL.md.disabled exists, stop before reading a fallback. Project configuration: If the current project contains .claude/skills/rip/SKILL.md, read and apply it instead of this bundled default. The project copy is the capability source of truth. Repository context: Read ../setup/references/task-context.md when resolving a task, project commands, named agent roles, or legacy .claude/ resource paths. Use repository evidence and applicable project instructions; a missing local workflow copy does not require setup. Resolve bundled resources from the installed skill, never from the target cwd.

Review Implementation Plan (RIP)

Step-by-step review of implementation plan. Focus: business value through technical understanding.

Resolve the plan and related inputs with ../setup/references/task-context.md. This is an optional read-only walkthrough after planning; it does not create a second plan or task.

Protocol

Phase 1: Analysis

Before responding:

1. Read the resolved active plan completely. If the argument is a directory or ambiguous path, use the task resolver and Glob for plan-shaped files (tech-decomposition-*.md, *implementation-plan*.md) before asking the user to choose.
2. Study related code (Grep, Read) — verify, don't assume. Never comment on a file you have not opened; if the plan names a file, read it before evaluating the step that touches it.
3. Compare with PRD/business requirements — run alignment checklist:
   • Measurable outcomes/KPIs defined?
   • Scope matches PRD (no creep, no gaps)?
   • Risks or dependencies identified?
   • Each step ties to a user-facing outcome?
4. Determine business value of each step
5. If an ambiguity or missing context would change the review, use AskUserQuestion to clarify; otherwise state the review assumption and continue.

Record mismatches (extra features, missed requirements, logic conflicts) for discussion. Frame each as facts + question rather than a verdict — the plan author knows the product context you don't, and a question invites a fix instead of a defense.

Review format:

## Overview

**Business goal:** [what problem it solves]
**Scope:** [components] | [files] | [complexity]
**Plan:** [N steps with names]

**Warnings (include uncertain ones with confidence tag):**
- [Problem] — contradicts [source]

Begin the walkthrough after the overview unless a material ambiguity needs user input.

Phase 2: Walkthrough

Default to a continuous walkthrough in one response (all steps, back-to-back) — 4.7 calibrates length well and one message is cheaper than N turns. Pause only when (a) the user asks you to pause, (b) you need to resolve a PRD mismatch before the next step makes sense, or (c) the plan has more than ~6 steps and a mid-review check-in improves comprehension.

Step format:

### Step [N]: [Name]

**Why:** [business reason]
**What:** [technical essence]
**Concepts:** [term]: [analogy]
**Impact:** [code] → [user-facing change]
**Code:** [if relevant]

**Warnings (include uncertain ones with confidence tag):**
- [PRD issue]
- **Question:** [clarification]

---

Phase 3: Summary

## Summary

**Covered:** [recap]
**Value:** [why it matters to user]
**Concepts:** [list]

**Self-check:**
1. [Business question]
2. [Technical question]

Rules

Example

Input: Progress tracking plan

Overview:

Business goal: Users see progress → motivation Scope: backend + DB | src/modules/progress/ | medium Plan: 4 steps (model, API, logic, integration)

Warning:

• Plan doesn't mention visualization — PRD requires charts. Out of scope?
• PRD: PDF export — not in plan. Include?

Step 1:

Why: Store word learning history What: WordProgress table links user+word, stores metrics Concepts: Database schema = DB blueprint as code Impact: "Progress" screen → word-level stats

Warning:

• PRD: track repetition intervals (spaced repetition). Schema: only lastSeen, no history.
• Question: Plan — current state only or full history?

Related Skills

Completion

The review is complete when every plan step has been walked through, each material mismatch or uncertainty is recorded with a confidence tag and a question, and the summary self-check is present. This is a read-only review; a warning or a user checkpoint does not authorize plan or code edits.