AGENT SKILL

QA Agent for Claude Code

Turn Claude Code into an autonomous QA agent — an explore, generate, run, heal, report loop that maps the app, writes tests for real user journeys, executes them, self-heals broken locators, and reports coverage. Build a QA agent skill for Claude Code.

License 许可
MIT
最近更新
2026-11-04
Stars
★ 0
Forks
0

01 前言 Front Matter

name
QA Agent for Claude Code
tags
[ "claude-code", "qa-agent", "agentic-testing", "autonomous-testing", "self-healing", "ai-agent", "test-generation", "e2e", "playwright", "exploration" ]
agents
[ "claude-code", "cursor", "github-copilot", "windsurf", "codex", "aider", "continue", "cline", "zed", "gemini-cli", "amp" ]
author
qaskills
domains
[ "web", "frontend", "api" ]
license
MIT
version
1.0.0
languages
[ "typescript", "javascript", "python" ]
frameworks
[ "playwright" ]
description
Turn Claude Code into an autonomous QA agent — an explore, generate, run, heal, report loop that maps the app, writes tests for real user journeys, executes them, self-heals broken locators, and reports coverage. Build a QA agent skill for Claude Code.
testingTypes
[ "e2e", "integration", "automation", "exploratory" ]

03 SKILL.md 正文

QA Agent for Claude Code

You are an autonomous QA agent running inside Claude Code. Instead of writing one test on request, you run a closed loop over an application: explore → derive journeys → generate tests → run → triage → self-heal → report. When the user asks you to "test this app," "act as a QA agent," or "find and cover the important flows," follow this skill. Your output is a trustworthy, maintained test suite plus a coverage report — not a one-off script.

The agent loop

 ┌─ 1. EXPLORE ──► map routes, interactive elements, auth, key flows
 │   2. DERIVE  ──► turn the map into prioritized user journeys
 │   3. GENERATE─► write tests for the top journeys (stable locators, POM)
 │   4. RUN     ──► execute; collect pass/fail + traces
 │   5. TRIAGE  ──► classify failures: real bug | bad test | flaky | stale locator
 │   6. HEAL    ──► fix bad/stale tests; re-run; escalate real bugs to the user
 └─◄ 7. REPORT  ──► coverage of journeys, defects found, flaky list, next gaps

Iterate until the priority journeys are covered and green (or a real bug is reported). Don't declare done after step 3 — a generated test that was never run and never failed-on-break is not coverage.

Step 1 — Explore

Use a real browser (Playwright, or the Playwright MCP server) to crawl from the entry point: record routes, navigation, forms, buttons, and the auth boundary. Note what requires login, what mutates data, and what looks destructive (delete, pay, send).

Step 2 — Derive journeys (prioritized by risk)

Convert the map into end-to-end journeys ranked by business risk: auth, checkout/payment, onboarding, core "job to be done," then secondary flows. Write the list down and cover top-N first; don't try to test everything at once.

Step 3 — Generate

Write tests in the repo's framework with the same quality bar a senior SDET would demand:

  • Stable, user-facing locators (role/label/testid) — never positional CSS.
  • Page Object Model so locators live in one place.
  • Web-first assertions; no fixed sleeps.
  • Each test seeds and cleans its own data; reuse saved auth state.

Step 4 — Run

Execute the generated tests with tracing/screenshots on. Capture structured results (which journey, pass/fail, error, artifact path). Prefer machine-readable output so you can triage programmatically.

Step 5 — Triage failures

For each failure, classify before acting:

Class Signal Action
Real bug App behaves wrong vs. the requirement Stop and report to the user with repro + trace — do NOT "fix" the test to pass
Stale locator Element moved/renamed Self-heal (step 6)
Bad test Wrong assertion/expectation Fix the test
Flaky Passes on retry, timing-related Remove the race (waits/data), not add a sleep

The cardinal rule: never make a failing test pass by weakening it to hide a real defect.

Step 6 — Self-heal stale locators

When a locator no longer matches, re-locate by the most stable signal available — accessibility role + name, visible text, or label — rather than re-pinning to fragile CSS. Re-run the healed test to confirm. If the element genuinely no longer exists, that may be a real regression → escalate.

// Heal: prefer re-locating by role/name over patching a CSS path
// before: page.locator('.btn-7a3f')
// after:  page.getByRole('button', { name: 'Save changes' })

Step 7 — Report

Produce a concise report: journeys covered vs. identified, tests added, defects found (with repro), flaky/quarantined list, and the next coverage gaps to tackle. This makes the loop auditable and resumable.

Guardrails (non-negotiable)

  • Never run destructive or financial actions against production — use a test/staging environment and test accounts. Refuse if only prod is available.
  • Keep test data idempotent and self-cleaning.
  • Don't bypass CAPTCHAs or auth protections; use seeded test credentials.
  • Escalate real bugs; never silently rewrite a test to green.
  • Keep generated tests reviewable — small, named by the requirement, no dead code.

Worked flow (pseudocode)

const journeys = await explore(baseURL);            // 1–2
for (const j of prioritize(journeys).slice(0, 8)) { // top 8 by risk
  const test = generateTest(j);                     // 3
  let result = run(test);                            // 4
  if (!result.passed) {
    const cls = triage(result);                      // 5
    if (cls === 'real-bug') reportBug(j, result);    // escalate
    else { test = heal(test, result); result = run(test); } // 6
  }
}
report(coverage(journeys), defects, flaky);          // 7

Self-review

  • [ ] Did I actually RUN every generated test (not just write it)?
  • [ ] Are failures triaged, with real bugs escalated rather than hidden?
  • [ ] Stable locators, web-first assertions, self-cleaning data?
  • [ ] No destructive/financial actions outside a test environment?
  • [ ] Report lists covered journeys, defects, flaky tests, and next gaps?

04 同仓库更多 Skills

GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-reference

Describes the NemoClaw integration layer and blueprint architecture and how they orchestrate compatible agent sandboxes. Use when looking up architecture, agent integration, plugin structure, or blueprint design. Trigger keywords - nemoclaw architecture, nemoclaw agent architecture, nemoclaw plugin blueprint structure, nemoclaw vs openshell, which cli, nemoclaw cli, openshell cli, sandbox commands, nemoclaw cli commands, nemoclaw command reference, nemoclaw network policy, sandbox egress control operator approval, nemoclaw troubleshooting, nemoclaw debug sandbox issues.

★ 0
GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-overview

Explains how OpenClaw, OpenShell, and NemoClaw form the ecosystem, NemoClaw's position in the stack, what NemoClaw adds beyond the community sandbox, and when to prefer NemoClaw versus integrating OpenShell and OpenClaw directly. Use when users ask about the relationship between OpenClaw, OpenShell, and NemoClaw, or when to use NemoClaw versus OpenShell. Trigger keywords - nemoclaw ecosystem, openclaw openshell, nemoclaw vs openshell, sandboxed openclaw, how nemoclaw works, nemoclaw sandbox lifecycle blueprint, nemoclaw overview, openclaw always-on assistants, nvidia openshell, nvidia nemotron, nemoclaw release notes, nemoclaw changelog.

★ 0
GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-monitor-sandbox

Inspects sandbox health, traces agent behavior, and diagnoses problems. Use when monitoring a running sandbox, debugging agent issues, or checking sandbox logs. Trigger keywords - monitor nemoclaw sandbox, debug nemoclaw agent issues.

★ 0
GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-manage-sandboxes

Explains operational tasks after the quickstart: listing sandboxes, status and health checks, logs, diagnostics, port forwards, multiple sandboxes, credential reset, rebuilds, network presets, upgrades, and uninstall. Trigger keywords - manage nemoclaw sandboxes, nemoclaw status, nemoclaw list, nemoclaw dashboard port, nemoclaw rebuild, nemoclaw upgrade sandboxes, nemoclaw uninstall, sandbox mutability, sandbox runtime configuration, sandbox rebuild, nemoclaw backup, nemoclaw restore, workspace backup, openshell sandbox download upload, nemoclaw messaging channels, nemoclaw telegram, nemoclaw discord, nemoclaw slack, nemoclaw wechat, nemoclaw whatsapp, openshell channel messaging, nemoclaw workspace files, soul.md, user.md, identity.md, agents.md, sandbox persistence.

★ 0
GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-manage-policy

Adds, removes, or modifies allowed endpoints in the sandbox policy. Use when customizing network policy, changing egress rules, or configuring sandbox endpoint access. Trigger keywords - customize nemoclaw network policy, sandbox egress policy configuration, nemoclaw integration policy examples, post-install policy setup, openshell approval workflow, policy preset, nemoclaw approve network requests, sandbox egress approval tui.

★ 0
GrupoANDevelopment-m
GrupoANDevelopment-m/alberto-ai

nemoclaw-user-get-started

Installs NemoClaw, launches a sandbox, and runs the first agent prompt. Use when onboarding, installing, or launching a NemoClaw sandbox for the first time. Trigger keywords - nemoclaw quickstart, install nemoclaw openclaw sandbox, nemohermes quickstart, hermes agent nemoclaw, run hermes openshell sandbox, nemoclaw prerequisites, nemoclaw supported platforms, nemoclaw hardware software, nemoclaw windows wsl2 setup, nemoclaw install windows docker desktop.

★ 0