Claude Agents para Engenharia de Qualidade
DHenry7471/claude_agents
Agentes Claude para testes, cobertura e quality gates em CI/CD
Ver no GitHubSuite completa de 12 agentes especializados para o workflow SDET, cobrindo design de estratégia de testes, geração de testes API/E2E com Playwright, análise de cobertura, triagem de falhas em CI e implementação de quality gates. Funciona em Claude Code, via npm ou como ferramenta MCP.
Instalação
npx convez-ai DHenry7471/claude_agents1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Desenhar estratégias de teste e auditar pirâmide de testes
- Gerar suites de testes API TypeScript com casos de erro e edge cases
- Escrever specs E2E em Playwright com Page Object Model
- Implementar quality gates em GitHub Actions, GitLab CI ou CircleCI
- Triar falhas de CI e classificar como regressão, flakiness ou ruído
- Analisar cobertura V8 e identificar gaps de lógica business critical
Veredito

Toolset maduro e opinionado para times SDET/QA que adotam Claude Code. Cobre o espectro completo do workflow de testes, mas requer familiaridade com TypeScript, Playwright e CI/CD para máximo valor.
README
A tradução para o português deste README ainda está na fila.
claude_agents
Twelve Claude Code subagents, five Horus API agents, and five skills covering the full SDET workflow — test strategy, API and Playwright E2E test generation, contract testing, test data factories, mutation analysis, coverage gap analysis, CI failure triage, and quality gate implementation.
Use them directly in Claude Code, invoke them programmatically via the npm package, or wire them into any MCP-compatible client as tools.
Two agent tiers:
- Standard agents (
agents/) — interactive, tool-enabled, for developers in the terminal - Horus agents (
agents/horus/) — API-only, JSON-in / JSON-out, for automation pipelines. Call withrunHorusAgent().
Agents
Custom subagents live in agents/. Each is a Markdown file with YAML frontmatter that Claude
Code loads and can invoke via the Agent tool or by name.
tessa-test-strategist
Designs comprehensive test strategies for features, services, or entire products. Audits
coverage across the full pyramid (unit, integration, E2E), identifies gaps, recommends tooling,
and produces a structured TEST_STRATEGY.md.
File: agents/tessa-test-strategist.md | Color: purple
ambrosine-api-tester
Generates TypeScript API test suites for REST or GraphQL endpoints. Covers happy paths, error cases, auth, schema validation, and edge cases. All external dependencies are mocked.
File: agents/ambrosine-api-tester.md | Color: green
ernie-e2e-test-writer
Writes production-ready Playwright E2E spec files for an existing test project. Reads project conventions, generates missing specs following the Page Object Model, and layers in axe-core accessibility tests.
File: agents/ernie-e2e-test-writer.md | Color: cyan
clint-ci-gatekeeper
Designs and implements CI/CD quality gates for GitHub Actions (or GitLab CI, CircleCI). Audits the existing pipeline against a full gate taxonomy and produces ready-to-commit workflow files.
File: agents/clint-ci-gatekeeper.md | Color: orange
felix-failure-triage
Triages CI test failures by classifying them as regressions, flaky tests, environment noise, or test bugs. Accepts Vitest and Playwright JSON reports plus flakiness history. Returns a structured verdict with root cause hypotheses, recommended owners, and a merge-block recommendation.
File: agents/felix-failure-triage.md | Color: red
greta-coverage-analyst
Analyses V8 coverage reports and source files to surface untested business logic ranked by risk — not just line percentage. Identifies uncovered error paths, state transitions, and validation branches. Produces a prioritised gap report with concrete test stubs for every CRITICAL and HIGH gap.
File: agents/greta-coverage-analyst.md | Color: yellow
iris-insight-reporter
Interprets CI run history data and produces a concise quality health summary for the Horus dashboard. Detects pass rate trends, coverage drift, pyramid imbalance, and anomalies. Returns a structured JSON payload (for programmatic consumption), an embeddable HTML snippet (for the dashboard panel), and a plain-text summary (for Slack or PR comments).
File: agents/iris-insight-reporter.md | Color: blue | Model: haiku
percy-pr-reviewer
Reviews pull request diffs touching test files and enforces ten quality engineering
standards: AAA pattern, given/when/then naming, pyramid layer compliance, mock
injection via @horus/test-utils, test isolation, no logic in tests, no hardcoded
waits, behavior-not-implementation assertions, assertion completeness, and test count
regression detection. Posts structured inline review comments with must-fix and
recommended categories.
File: agents/percy-pr-reviewer.md | Color: pink | Model: haiku | Tools: Read, Glob, Grep
saxon-spec-to-test
Converts ADRs, GitHub issues, feature specs, or user stories into production-ready Vitest test scaffolds. Reads the codebase to use correct method signatures, types, and builders. Assigns each scenario to the right pyramid layer and generates complete files with AAA stubs ready to fill in.
File: agents/saxon-spec-to-test.md | Color: purple
pat-pact-contract-tester
Designs and implements consumer-driven contract tests using Pact for microservice boundaries.
Generates consumer pact files, provider verification tests, and a Pact Broker CI pipeline with
can-i-deploy gates. Identifies which service boundaries lack contract coverage and are at risk
of silent breakage.
File: agents/pat-pact-contract-tester.md | Color: purple
furio-forge-test-data
Generates realistic, schema-compliant test fixtures and typed builder factories. Reads TypeScript
types, Zod/Joi schemas, Prisma models, or OpenAPI specs and produces a faker-backed builder
library with sensible defaults and per-test override support. Eliminates raw object literals
scattered across test files.
File: agents/furio-forge-test-data.md | Color: orange
kurt-striker-mutation-analyst
Interprets Stryker mutation testing reports, triages surviving mutants by risk (CRITICAL / HIGH / LOW), and generates the minimal set of targeted tests needed to kill the most dangerous survivors. Distinguishes high-value kills (business logic, auth, validation) from acceptable survivors (logging, trivial getters). Projects the new mutation score after kills.
File: agents/kurt-striker-mutation-analyst.md | Color: red
Horus Agents
Five agents in agents/horus/ are designed for Horus automation pipelines. They require no
tool access — the calling code pre-fetches all data and passes it as a JSON-serialised input
object. They always return a single JSON code block.
| Agent | Short alias | Contract |
|---|---|---|
horus-felix-failure-triage | horus-felix | FelixInput → FelixOutput |
horus-greta-coverage-analyst | horus-greta | GretaInput → GretaOutput |
horus-iris-insight-reporter | horus-iris | IrisInput → IrisOutput |
horus-percy-pr-reviewer | horus-percy | PercyInput → PercyOutput |
horus-kurt-striker-mutation-analyst | horus-kurt | KurtInput → KurtOutput |
TypeScript types for all inputs and outputs live in shared/contracts/. The InsightRecord
persistence schema lives in shared/insight-store/.
Adding a New Agent
Agents are plain Markdown files with YAML frontmatter stored in agents/.
1. Create the agent file
touch agents/my-agent.md
2. Write the frontmatter and instructions
---
name: my-agent
description: >
One or two sentences describing what this agent does and when to invoke it.
Be specific — Claude Code uses this text to decide when to route to the agent.
model: inherit
color: blue
tools: ["Read", "Write", "Edit", "Bash", "Glob", "Grep"]
---
You are a [role]. Your job is to [goal].
## Phase 1 — ...
Frontmatter fields:
| Field | Required | Notes |
|---|---|---|
name | Yes | Lowercase, hyphens only, max 64 chars |
description | Yes | Used for routing; be specific, max 1024 chars |
model | No | inherit uses the current session model; or pin a model ID |
color | No | red, orange, yellow, green, cyan, blue, purple, pink |
tools | No | Restrict tool access; omit to allow all tools |
3. Install in Claude Code
User-level (available in all projects):
cp agents/my-agent.md ~/.claude/agents/
Project-level (this repo only):
cp agents/my-agent.md .claude/agents/
Claude Code auto-discovers agents from both locations on startup.
Skills
All skills live in skills/testing/ and follow the Agent Skills standard.
test-architect
Designs comprehensive test strategies for services, features, or entire products. Produces a structured test plan covering the full pyramid (unit, integration, E2E), identifies coverage gaps, recommends tooling, and defines success criteria.
Invoke: /skill:test-architect
Install (Claude desktop): skills/testing/test-architect.skill
api-test-engineer
Generates REST/GraphQL API test suites in TypeScript. Covers happy paths, error cases, auth, schema validation, and edge cases. All external dependencies are mocked — no live services.
Invoke: /skill:api-test-engineer
Install (Claude desktop): skills/testing/api-test-engineer.skill
accessibility-auditor
Audits web pages for WCAG 2.1/2.2 AA violations using axe-core with Playwright. Produces a
prioritized violation report and ready-to-run accessibility.spec.ts.
Invoke: /skill:accessibility-auditor
Install (Claude desktop): skills/testing/accessibility-auditor.skill
ci-quality-gatekeeper
Designs and implements CI/CD quality gates for GitHub Actions (or GitLab CI, CircleCI). Covers lint, type check, unit coverage thresholds, integration tests, E2E, security scanning, and accessibility — all blocking, not just warning.
Invoke: /skill:ci-quality-gatekeeper
Install (Claude desktop): skills/testing/ci-quality-gatekeeper.skill
playwright-qa-agent
Full-site Playwright UI QA agent. Crawls a site, discovers all significant pages and user flows, generates a complete multi-page TypeScript test suite with Page Objects, executes the tests, and outputs a GitHub Actions CI config.
Invoke: /skill:playwright-qa-agent
Install (Claude desktop): Save skills/testing/playwright-qa-agent.skill via the Claude desktop app.
Adding a New Skill
Skills are plain directories with a SKILL.md file. Both Claude and Pi discover and load them
automatically.
1. Create the skill directory
mkdir -p skills/testing/my-skill
2. Write SKILL.md
---
name: my-skill
description: >
One or two sentences describing what this skill does and when to use it.
Be specific — this text determines when the agent loads the skill.
---
# My Skill
Instructions for the agent go here. Include setup steps, usage patterns,
expected outputs, and any scripts or reference files to load.
Frontmatter rules: name must be lowercase with hyphens only, max 64 chars. description is
required and max 1024 chars. See the Agent Skills spec
for the full reference.
3. Install in Claude desktop
# Zip the skill directory into a .skill bundle
cd skills/testing
zip -r my-skill.skill my-skill/
Then drag my-skill.skill into the Claude desktop app (Settings → Skills → Install from file),
or save it via the skills panel.
4. Use in Pi
No extra steps — Pi auto-discovers skills from skills/testing/ via .pi/settings.json. Restart
Pi and invoke with /skill:my-skill.
npm Package
The mcp/ directory is published as @wutangbanger/claude-agents — a self-contained package with all agent and skill prompts bundled at build time. No file paths or separate repo checkout required.
Programmatic use
npm install @wutangbanger/claude-agents
# or
pnpm add @wutangbanger/claude-agents
Standard agents — returns markdown prose in output:
import { runAgent } from '@wutangbanger/claude-agents';
const { output } = await runAgent('felix-failure-triage', task);
const { output } = await runAgent('felix', task); // short alias
Horus agents — accepts a typed input object, returns parsed JSON in data:
import { runHorusAgent } from '@wutangbanger/claude-agents';
import type { FelixInput, FelixOutput } from '@wutangbanger/claude-agents/contracts';
const input: FelixInput = {
ciReport: vitestJsonReport,
gitDiff: 'src/orders/orderService.ts\ntests/orders/orderService.test.ts',
branch: 'feat/payment-refactor',
runId: '12345',
};
const { data } = await runHorusAgent<FelixOutput>('horus-felix', input);
// data.mergeRecommendation === 'BLOCK' | 'ALLOW'
// data.failures[0].classification === 'REGRESSION' | 'FLAKY' | ...
Reads ANTHROPIC_API_KEY from the environment. Override per-call via options.apiKey.
MCP server (Claude Code / IDE integration)
ANTHROPIC_API_KEY=sk-... npx @wutangbanger/claude-agents
Or add to .mcp.json / claude_desktop_config.json:
{
"mcpServers": {
"claude-agents": {
"command": "npx",
"args": ["@wutangbanger/claude-agents"],
"env": { "ANTHROPIC_API_KEY": "sk-..." }
}
}
}
Registered tools
Standard agents and skills accept task (string), model?, max_tokens?.
Horus agents accept input (JSON string per contract), model?, max_tokens?.
| Tool | Type |
|---|---|
tessa-test-strategist | Agent |
ambrosine-api-tester | Agent |
ernie-e2e-test-writer | Agent |
clint-ci-gatekeeper | Agent |
felix-failure-triage | Agent |
greta-coverage-analyst | Agent |
iris-insight-reporter | Agent |
percy-pr-reviewer | Agent |
saxon-spec-to-test | Agent |
pat-pact-contract-tester | Agent |
furio-forge-test-data | Agent |
kurt-striker-mutation-analyst | Agent |
horus-felix-failure-triage | Horus Agent |
horus-greta-coverage-analyst | Horus Agent |
horus-iris-insight-reporter | Horus Agent |
horus-percy-pr-reviewer | Horus Agent |
horus-kurt-striker-mutation-analyst | Horus Agent |
skill-testing-test-architect | Skill |
skill-testing-api-test-engineer | Skill |
skill-testing-accessibility-auditor | Skill |
skill-testing-ci-quality-gatekeeper | Skill |
skill-testing-playwright-qa-agent | Skill |
list-agents-and-skills | Utility |
Adding or editing agents/skills requires rebuilding and republishing the package (pnpm run build && pnpm publish --access public from mcp/).
Pi Integration
This repo supports Pi coding agent out of the box.
npm install -g --ignore-scripts @earendil-works/pi-coding-agent
export ANTHROPIC_API_KEY=your_key
cd claude_agents
pi
All skills are auto-discovered from skills/testing/ via .pi/settings.json. Invoke any skill
with /skill:<name>.
Structure
.pi/
settings.json # Pi config
AGENTS.md # Repo context for Pi and other coding agents
agents/ # Standard Claude Code subagents (interactive, tool-enabled)
tessa-test-strategist.md # Test strategy designer
ambrosine-api-tester.md # API test suite generator
ernie-e2e-test-writer.md # Playwright E2E spec writer
clint-ci-gatekeeper.md # CI/CD quality gate implementer
felix-failure-triage.md # CI failure classifier and triage reporter
greta-coverage-analyst.md # Risk-ranked coverage gap analyst
iris-insight-reporter.md # CI history quality health reporter
percy-pr-reviewer.md # Test code PR reviewer (10 standards enforced)
saxon-spec-to-test.md # Spec-to-test scaffold generator
pat-pact-contract-tester.md # Consumer-driven contract tests via Pact
furio-forge-test-data.md # Test fixture and builder factory generator
kurt-striker-mutation-analyst.md # Stryker mutation report analyst
horus/ # Horus API variants (JSON-in/JSON-out, no tool calls)
felix-failure-triage.md # FelixInput → FelixOutput
greta-coverage-analyst.md # GretaInput → GretaOutput
iris-insight-reporter.md # IrisInput → IrisOutput
percy-pr-reviewer.md # PercyInput → PercyOutput
kurt-striker-mutation-analyst.md # KurtInput → KurtOutput
shared/ # TypeScript contracts — source of truth for Horus I/O types
contracts/
felix.ts # FelixInput, FelixOutput
greta.ts # GretaInput, GretaOutput
iris.ts # IrisInput, IrisOutput
percy.ts # PercyInput, PercyOutput
kurt.ts # KurtInput, KurtOutput
index.ts # barrel export
insight-store/
schema.ts # InsightRecord, InsightStore, InsightQuery
index.ts # barrel export
mcp/ # Published as @wutangbanger/claude-agents
scripts/
bundle-prompts.mjs # Prebuild: reads agents/ + agents/horus/ + skills/
src/
index.ts # Exports runAgent(), runHorusAgent(), listHorusAgents(), types
api.ts # runAgent() + runHorusAgent() (Anthropic SDK)
registry.ts # Lookup/list + SLUG_ALIASES (includes horus-* aliases)
types.ts # AgentDef (horus?: bool), AgentResult, HorusAgentResult, …
mcp.ts # MCP server — registers standard + horus + skill tools
generated/ # Auto-generated by bundle-prompts.mjs — gitignored
package.json # version 3.2.0
tsconfig.json
README.md
skills/
testing/
test-architect/
api-test-engineer/
accessibility-auditor/
ci-quality-gatekeeper/
playwright-qa-agent/