review-agent

Independently reviews built slices against design files, data stores, and done criteria.

Updated Apr 13, 2026
One-click install
npx skills add https://github.com/scoots31/engineering-playbook --skill review-agent-scoots31
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: review-agent
Source: https://github.com/scoots31/engineering-playbook/tree/main/skills/review-agent
Command: npx skills add https://github.com/scoots31/engineering-playbook --skill review-agent-scoots31

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Builders cannot neutrally judge their own work, so self-reported QA often misses gaps between the design spec and the running application. This Skill provides an independent review pass that verifies every claim through direct observation. ## Core Features & Use Cases - Independent observation: Reads the backlog slice record, design file, and data spec directly, never consulting the builder's QA manifest or self-report. - Full verification pass: Screenshots the running app, checks console and network errors, tests interactions, resizes to mobile, and reads the data store directly to confirm written data. - Binary verdicts: Returns only CLEARED or GAPS with specific expected-vs-observed entries for every done criterion, design data point, and quality contract item. - Use Case: After a builder signals QA complete on a slice, run this review to confirm the login form matches the design file, the user record actually exists in the database, and all done criteria pass before advancing the slice to In Test. ## Quick Start Ask the review agent to independently review slice SL-12 using the backlog, design file, data spec, and review URL, and return a CLEARED or GAPS verdict.

Frequently Asked Questions about review-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I independently review a built slice before marking it done?

Provide the slice ID, backlog path, design file path, data spec path, and review URL. The review reads all sources directly, screenshots the running app, reads the data store, and returns CLEARED or GAPS with specific observations for each criterion.

What does the review agent check during a slice review?

It verifies every data-sourced design element on screen, console errors, failed network requests, interactive element behavior, mobile rendering at 375x812, direct data store reads, all done criteria, and observable quality contract items.

Can the review agent run in Cursor without an agent tool?

Yes, it runs inline within the same context in Cursor. The same independence standard applies: it reads the design file and data spec fresh, ignores prior build observations, and produces the identical CLEARED or GAPS output format.

What happens when the review returns GAPS?

The builder receives the specific gap list, addresses every gap, reruns the full QA pass, and signals completion again. The review then reruns from scratch. If the same gap appears three consecutive times, it is surfaced to the solo for a decision.

Does the review agent accept the builder's QA manifest as evidence?

No. It never receives or consults the builder's QA manifest or self-report. Every conclusion comes from direct observation: screenshots of the running output and direct reads of the data store, never claims about what was written.