home / skills / shipshitdev / library / qa-reviewer

qa-reviewer skill

safe

This skill helps you verify AI outputs for accuracy, completeness, and alignment with requirements before committing changes.

npx playbooks add skill shipshitdev/library --skill qa-reviewer

Review the files below or copy the command above to add this skill to your agents.

Files (3)

SKILL.md

2.4 KB

---
name: qa-reviewer
description: Systematically review AI agent work for quality, accuracy, and completeness. Catches bugs, verifies patterns, checks against requirements, and suggests improvements before committing changes.
version: 1.0.0
tags:
  - quality-assurance
  - verification
  - code-review
  - accuracy
  - completeness
---

# QA Reviewer: Systematic Work Verification

## Purpose

Structured framework for reviewing AI agent work before finalizing changes. Catches bugs, verifies accuracy, ensures completeness, validates solutions match requirements.

**Why this exists:** AI agents can introduce subtle bugs, miss requirements, or make incorrect assumptions.

## When to Use

- User says "check your work" or "review this"
- After completing complex multi-step implementations
- Before committing major refactors
- When modifying commands or skills
- After implementing new features
- Proactively after any task >5 steps

## QA Workflow Phases

### 1. Context Gathering

- Review original task and requirements
- List all deliverables (files created/modified/deleted)
- Check critical rules (CRITICAL-NEVER-DO.md)

### 2. Requirement Verification

- Create checklist of all requirements
- Verify solution alignment
- Check for edge cases

### 3. Accuracy Verification

- Verify file paths exist
- Confirm patterns match codebase
- Cross-reference citations

### 4. Bug Detection

- Check syntax (markdown, code, paths)
- Look for logic errors
- Verify project-specific rules

### 5. Completeness Audit

- Reconcile todo list
- Verify all mentioned files changed
- Check for side effects

### 6. Optimization Review

- Evaluate solution quality
- Check pattern consistency
- Consider alternatives

## Quick Verification Commands

```bash
# Verify file exists
ls -la <file-path>

# Check balanced code blocks
grep -c '\`\`\`' file.md

# Find violations
grep -r "console\.log" <files>
grep -r ": any" <files>
```

## Project Rules Check

- [ ] No console.log (use logger)
- [ ] No `any` types
- [ ] No inline interfaces
- [ ] AGENTS/CLAUDE/CODEX files intact
- [ ] .agents/ folders untouched
- [ ] Multi-tenancy preserved

## Final Assessment Categories

- **Approve** - Ready to commit
- **Conditional** - Fix issues first
- **Reject** - Significant problems

---

**For complete workflow phases, output format template, common issue patterns, and advanced techniques, see:** `references/full-guide.md`

Overview

This skill systematically reviews AI agent work for quality, accuracy, and completeness before changes are finalized. It catches bugs, verifies patterns, checks requirements, and suggests concrete fixes to make code and documentation production-ready. The workflow is optimized for agent-driven projects and command/skill changes.

How this skill works

The skill inspects task context, lists deliverables, and builds a requirements checklist to verify alignment. It runs targeted checks: file and path validation, syntax and pattern scans, logic and edge-case reviews, and project-rule enforcement. Results are categorized as Approve, Conditional, or Reject with actionable remediation steps.

When to use it

When a user requests “check your work” or “review this”
After completing complex multi-step implementations
Before committing major refactors or feature merges
When modifying commands, skills, or agent behaviors
Proactively after any task with more than five steps

Best practices

Start by collecting the original task and explicit requirements; create a concise checklist
List all files created, modified, or deleted to reconcile changes
Run pattern and rule scans specific to the project (e.g., banned patterns, type rules)
Prioritize reproducible checks: path existence, balanced code blocks, and basic linting
Classify findings as Approve/Conditional/Reject and include specific remediation steps

Example use cases

Review a multi-file feature implementation for missing edge cases and side effects
Validate that a refactor did not introduce banned patterns or break multi-tenancy rules
Check new agent commands for correct file paths, syntax, and adherence to project conventions
Audit a pull request to produce a concise pass/fail assessment and a TODO remediation list
Perform a pre-commit check focused on project-specific rules (no console.log, no any types)

FAQ

What does a Conditional assessment mean?

Conditional means work is close but requires specific fixes before approval; the report lists required changes and verification steps.

Which quick checks are recommended first?

Start with file/path existence, balanced code block counts, and simple pattern searches for banned tokens (e.g., console.log, any types) to catch high-impact issues quickly.