Stop writing ad-hoc prompts. Start using engineered AI skills for QA.

This repository is a ready-to-use library of 21 AI skills (reusable prompts that tell the AI exactly how to perform a QA task), 31 anti-pattern quality gates (rules the AI checks its own output against before finishing), and 3 specialized agents (AI roles — one writes test code, one reviews it, one generates load tests) designed specifically for QA workflows. Copy the .claude/ folder into your project, and your AI assistant immediately knows how to audit specs, generate test cases, write API tests (Kotlin/Java), and check its own output.

Works with Claude Code, OpenCode, Cursor, VS Code Copilot, Codex, JetBrains AI, and Gemini Code Assist.

AI coding assistants can write your API tests, audit your specs, and review your code — but only if you give them the right instructions. Without context, the AI starts from zero every conversation: it picks random libraries, ignores your coding standards, and produces output you have to rewrite. This repository contains field-tested instructions that make the AI behave like a senior QA engineer who already knows your project.

You don't need to read the whole repo. Three steps to start getting value:

1. Copy — Copy the .claude/ folder from this repo into your backend or QA project root. For non-Claude IDEs (Cursor, Copilot, etc.), see IDE Compatibility below.

2. Create CLAUDE.md by hand — Write a minimal CLAUDE.md in your project root with only: tech stack, build/test commands, and banned alternatives. See docs/claudemd-instructions.md for guidelines. Research shows that hand-written context files outperform AI-generated ones (+4% vs −3% success rate).

3. Run your first skill — Try scanning a backend repo:

   /repo-scout
   

   The AI reads your codebase and produces a structured report: API endpoints found, tech stack detected, test coverage gaps, and a blueprint for test generation.

New to AI coding assistants? Start with the spec-only branch — it contains only API specifications with no pre-generated code, so you can follow the full pipeline from scratch. See Demo Video for a walkthrough.

The main pipeline — choose your starting point based on scope:

/repo-scout       →  /api-test-cases     →  /api-tests      →  /api-test-review
(backend repo)       (test scenarios)       (QA test repo)     (code review)

While /api-tests provides good coverage out of the box, it is designed to be adapted to your team's architectural guidelines.

Optional: /api-isolated-tests — generates detailed test scenarios for a single endpoint (steps, data, expected results). Use when you need a deep-dive into one area instead of full API coverage.

Here is what real skill output looks like (truncated). Each example links to the full artifact in this repo.

/spec-audit — finds contradictions in API specifications:

Spec Audit — Registration API v1
Verdict: BLOCKED | Score: 0%

Top 3 Risks:
1. [BLOCKER] Spec declares 2FA via SMS but phone field is absent from Request Body
2. [CRITICAL] No HTTP Responses section — no success code, no error codes
3. [CRITICAL] Example payload violates Business Rule 3 (password contains "Alex")

Full report: audit/spec-audit_registration-api-v1_20260226_000000.md

/api-test-cases — generates exhaustive test scenario matrix:

Feature: User Registration (POST /api/v1/users/register) [CRITICAL]

| ID     | Type | Scenario                          | Expected Result          |
|--------|------|-----------------------------------|--------------------------|
| REG-01 | POS  | Happy path — minimal valid data   | 201 Created + JWT token  |
| REG-07 | NEG  | Missing email field               | 400 + VALIDATION_ERROR   |
| REG-12 | NEG  | Email is empty string             | 400 + VALIDATION_ERROR   |
| ...    |      | (40+ scenarios per endpoint)      |                          |

Full scenarios: docs/test-cases/test-scenarios_20260226_120000.md

/api-test-review — deep code review of generated tests:

API Test Review Report: Registration
Scope: 19 files, 70+ tests

🔴 CRITICAL: Thread.sleep(2000) in RegistrationIdempotencyTests.kt:45
   Fix: Replace with runTest { advanceTimeBy(2.seconds) }

🟠 MAJOR: Missing Content-Type assertion in RegistrationPositiveTests.kt:28
   Fix: response.contentType() shouldBe ContentType.Application.Json

✅ Security: No issues found
✅ Allure Integration: No issues found

Full review: audit/api-test-review-report_registration_20260301_143000.md

When a test suite already exists, skip generation and focus on auditing and remediating:

/repo-scout  →  /api-test-cases  →  /api-test-review  →  /api-tests fix
(fresh map)     (gap analysis)      (legacy audit +        (surgical fixes,
                                     contract check)         no regeneration)

Key differences from the greenfield flow:

• Skip /api-tests generate — tests already exist
• /api-test-review reads API contracts (Swagger/OpenAPI/Protobuf/GraphQL) from the repo and validates test DTOs against the actual spec
• /api-tests fix automatically fixes common issues (Thread.sleep, runBlocking, missing timeouts) without rewriting test logic

Full catalog of all 21 skills (setup, audit, analysis, translation): docs/ai-setup.md

Disclaimer: Always review AI-generated results. Even with well-crafted prompts and agents, outputs must be validated by a human before being merged or executed.

Research shows that bloated or LLM-generated context files reduce agent success rate and increase inference cost by over 20%. Keep your CLAUDE.md as small as possible: only specific tooling, build commands, and banned alternatives. Do not add codebase overviews or duplicate existing documentation.

See docs/claudemd-instructions.md for the full guidelines.

This library is a starting point. To make it yours:

1. Run a skill and review. Expect some gaps on the first try.
2. Tweak the .md files. Skills are just natural language. Add your team's specific requirements, remove noise, or ask the AI to improve the prompt directly.
3. Iterate and share. After major edits, run /skill-audit to check quality.
4. Once results are consistently good, share the updated skill files with your team.

IDE-specific prompts for running each skill: docs/workshop-commands.md

• 21 skills in .claude/skills/ — from repo scanning to test generation to translation
• 31 anti-pattern quality gates in .claude/qa-antipatterns/ — the AI checks generated code against these before finishing
• 3 specialized agents in .claude/agents/ — SDET (writes test code), Auditor (reviews quality), and Perf Engineer (generates load tests)
• Layered context loading — the AI reads only what it needs for the current task (CLAUDE.md always, agent/skill files on demand), keeping responses fast and focused
• Self-improving loop — the AI suggests improvements to its own knowledge base at the end of each run, so skills get better over time
• Chained pipeline — each skill builds on the previous one's output (/repo-scout → /api-test-cases → /api-tests → /api-test-review), so results are consistent and traceable

Full inventory of all files and architectural patterns: docs/ai-setup.md

Kotlin (default — /api-tests):

Java 17+ (opt-in — /api-tests-java):

• Demo Video — capability walkthrough, presented at Podlodka AI Crew #2 (February 2026)
• Presentation (PDF)
• Workshop commands & IDE prompts
• Branches: main (fully configured project with generated tests), spec-only (clean starting point — API specs only, no generated code; best for first-time users)