vibe-devil-advocate-review
npx machina-cli add skill ash1794/vibe-engineering/devil-advocate-review --openclawFiles (1)
SKILL.md
2.1 KB
vibe-devil-advocate-review
Before shipping a recommendation, challenge it. The goal is to find weaknesses BEFORE they become production incidents.
When to Use This Skill
- Before sending a design document for approval
- Before shipping a recommendation that combines multiple inputs
- Before merging a large feature branch
- When you feel "too confident" about a solution
- User asks for a review or second opinion
When NOT to Use This Skill
- Trivial changes (typo fixes, formatting)
- When the user explicitly says "just ship it"
- During brainstorming (don't kill ideas before they form)
- For code review of small PRs (use
vibe-quality-loopinstead)
The 5 Dimensions
Challenge the work across:
- Consistency — Do all parts agree with each other? Are there contradictions?
- Completeness — What's missing? What edge cases are unaddressed? What blind spots exist?
- Actionability — Is every recommendation concrete and measurable? Can someone actually do this?
- Alignment — Does this match the stated goals, constraints, and user needs?
- Risk — What could go wrong? What are the second-order effects? What's the blast radius of failure?
Steps
- Read the artifact thoroughly — don't skim
- For each dimension, actively try to find problems. Assume there ARE problems.
- Score each dimension 1-5 (1=critical issues, 5=solid)
- List specific issues with evidence
- Deliver verdict: APPROVE / REVISE (with required changes) / REJECT (with blocking issues)
Output Format
Devil's Advocate Review
| Dimension | Score | Issues Found |
|---|---|---|
| Consistency | X/5 | [count] |
| Completeness | X/5 | [count] |
| Actionability | X/5 | [count] |
| Alignment | X/5 | [count] |
| Risk | X/5 | [count] |
Critical Issues
- [Issue with evidence]
Warnings
- [Non-blocking concern]
Verdict: APPROVE / REVISE / REJECT
[Rationale]
Source
git clone https://github.com/ash1794/vibe-engineering/blob/master/skills/devil-advocate-review/SKILL.mdView on GitHub Overview
Challenge recommendations, designs, or syntheses across five dimensions to surface weaknesses before production incidents. It helps teams uncover contradictions, gaps, and risks early, especially before approvals or large feature merges.
How This Skill Works
Read the artifact thoroughly, then for each of the five dimensions (Consistency, Completeness, Actionability, Alignment, Risk) score 1-5. List specific issues with evidence and deliver a verdict: APPROVE / REVISE (with required changes) / REJECT (with blocking issues). Capture issues per dimension and provide a clear rationale for the final verdict.
When to Use It
- Before sending a design document for approval
- Before shipping a recommendation that combines multiple inputs
- Before merging a large feature branch
- When you feel "too confident" about a solution
- User asks for a review or second opinion
Quick Start
- Step 1: Read the artifact thoroughly — don’t skim.
- Step 2: Score each of the five dimensions (1-5) and note evidence.
- Step 3: Deliver verdict and list required changes.
Best Practices
- Read the artifact thoroughly; don’t skim.
- Actively test each of the five dimensions for contradictions, gaps, and risks.
- Score each dimension 1-5 and attach concrete evidence for each point.
- Document concrete issues with evidence and organize them by dimension.
- Deliver a clear verdict (APPROVE / REVISE / REJECT) with required changes.
Example Use Cases
- Review a product design doc for a new feature to catch inconsistent assumptions.
- Adversarially review a data-sourcing recommendation that combines inputs from multiple teams.
- Goverance review of a large architectural decision before a major merge.
- Second-opinion check for a high-risk deployment plan with rollback considerations.
- Advising on a cross-team synthesis doc before rollout to multiple stochastic environments.
Frequently Asked Questions
Add this skill to your agents
