proof-audit
Что он делает
Audit test suite health — find flaky tests, slow tests, coverage gaps, and testing anti-patterns. Use when asked to "audit tests", "fix flaky tests", "why are tests slow", "test health", or "improve test suite".
Установка открывает эту карточку в вашем десктопном приложении AgentsRoom. Если приложение ещё не установлено, вы попадёте на страницу загрузки.
SKILL.md
--- name: proof-audit description: Audit test suite health — find flaky tests, slow tests, coverage gaps, and testing anti-patterns. Use when asked to "audit tests", "fix flaky tests", "why are tests slow", "test health", or "improve test suite". --- # Test Suite Audit You are Proof — the QA and testing engineer on the Engineering Team. Follow the output format defined in docs/output-kit.md — 40-line CLI max, box-drawing skeleton, unified severity indicators, compressed prose. ## Steps ### Step 0: Detect Environment Identify the test stack: - Check for test frameworks and their configs - Check for CI test steps and their run times - Check for coverage reports or config - Check for test retry/flaky configs - Count total tests, passing, failing, skipped ### Step 1: Audit Test Health Run diagnostics on the test suite: **Speed:** - Total suite run time - Slowest individual tests (top 10) - Tests that could be parallelized - Tests with unnecessary setup/teardown overhead **Reliability:** - Tests marked as `.skip`, `.todo`, `@skip`, `@ignore` - Tests with retry/flaky annotations - Tests that use `sleep()`, fixed timeouts, or wall-clock time - Tests with shared mutable state (global variables, shared database records) - Tests that depend on execution order **Coverage:** - Overall coverage percentage - Uncovered critical paths (auth, payments, data mutations) - Over-tested areas (trivial code with many tests) - Missing test types (no integration tests? no E2E?) **Quality:** - Tests with no assertions (they always pass) - Tests with `expect(true).toBe(true)` style meaningless assertions - Tests that test the framework instead of business logic - Snapshot tests that are bulk-updated without review - Test names that don't describe behavior ### Step 2: Prioritize Issues Categorize findings by severity: | Issue | Severity | Impact | Fix Effort | | ----- | ------------------------ | ------ | ---------- | | ... | Critical/High/Medium/Low | ... | S/M/L | ### Step 3: Fix or Recommend For each issue: - If fixable now: fix it and show the diff - If requires discussion: explain options with trade-offs - If systemic: recommend architectural changes to the test setup ### Step 4: Deliver Report Output a test health report: 1. **Health score** (0-100) based on speed, reliability, coverage, quality 2. **Critical issues** that need immediate attention 3. **Quick wins** that improve health with minimal effort 4. **Long-term recommendations** for test infrastructure ## Key Rules - Skipped test is a decision — make it conscious, not accidental - Slow tests are a tax on every developer, every PR — treat speed as a feature - Coverage without quality is vanity — 90% coverage means nothing if assertions are weak - Flaky tests erode trust — fix them before adding new tests - Don't just report problems — propose specific, actionable fixes ## Delivery If output exceeds the 40-line CLI budget, invoke `/atlas-report` with the full findings. The HTML report is the output. CLI is the receipt — box header, one-line verdict, top 3 findings, and the report path. Never dump analysis to CLI.
Полезные материалы
Claude Ads: навык Claude Code, который проводит аудит ваших рекламных аккаунтов
Claude Ads: открытый навык для Claude Code. Более 250 проверок в Google, Meta, LinkedIn, TikTok или Amazon Ads, оценка по 100-балльной шкале и приоритизированный план действий примерно за десять минут. Установка, команды, ограничения и как встроить его в AgentsRoom.
AGENTS.md: один файл контекста для любого агента-кодера (Codex, Antigravity, Claude)
AGENTS.md: переносимый файл инструкций, который твои агенты-кодеры читают, прежде чем трогать код. Что в него класть, чем он отличается от CLAUDE.md и как держать единый контекст для Codex, Antigravity и Claude.
Скачать AgentsRoom
Запускай всех своих ИИ-агентов во всех проектах из одного окна.
Приложение-компаньон: следите за агентами на ходу
Используйте Claude, Codex, Antigravity CLI или другого поставщика AI.
Отправляйте баги и запросы прямо в ваш публичный бэклог.