Autonomous QA Orchestration
Asked to autonomously test an application end-to-end with minimal hand-holding — discover, explore, plan, execute, investigate, report, and verify fixes as one continuous QA pass. The top-level playbook that pulls in the other testing skills as needed.
The playbook for an open-ended request like "go test this app". It sequences the other skills; load each one when you reach its stage. Work in a time-boxed session with a written charter, stay within the authorized scope, and finish with a structured report.
Before you start
- Scope and authorization: which app/environment (prefer staging or local), which accounts, what is off-limits (production data, payments, third parties, destructive actions). If unclear, ask once.
- Charter and time-box: goal, areas, budget (for example 60–90 minutes). Say what you will not cover.
- Safety: synthetic data only; sandbox payment credentials; no load, fuzz or security scanning without explicit permission (
security-testing).
The pass
- Discover the application: purpose, stack, environments, existing docs, past reports and known issues.
- Understand the architecture: frontend/backend split, API shape, datastore, integrations, auth model. Skim; find where risk concentrates.
- Understand what changed:
git log/diff, linked tickets/PRD. For a general audit, prioritize by risk (test-strategist,qa-fundamentals). - Explore the UI: landmark and feature tours (
exploratory-testing); use a live browser tool (Playwright, Chrome DevTools MCP, or similar) when available (qa-test-harness). - Explore the API: capture real requests from the network tab or spec and probe them (
api-testing). - Plan: a short risk-based plan from what you actually found (
test-planning-documentation), not a generic checklist. - Design cases: use techniques (
test-design-techniques) and any spec/schema for generation (qa-code-intelligence). - Execute: exploratory where unknown, automated where repeatable (
test-automation). - Reproduce bugs minimally before filing; note the reproduction rate.
- Collect evidence: screenshots, console and network logs, request/response pairs, relevant data state.
- Trace root cause through logs/traces where you have access (
observability-testing). - Security quick-pass on auth, inputs and data boundaries (
security-testing,authn-authz-testing). - Performance and accessibility spot checks on key pages (
web-performance-testing,accessibility-testing). - Regression check against baseline or the last deploy (
regression-testing). - Report once, structured (
bug-reporting,test-metrics-reporting). - File tickets in the team's tracker if you have access (GitHub/Jira CLI or MCP) rather than leaving findings only in chat.
- Verify fixes by re-running the exact original repro.
- Keep the value: promote high-value bugs to permanent regression tests.
Operating discipline
- Never declare "passed" on appearances: verify status codes, DOM/state and persisted data.
- State assumptions explicitly when requirements are missing; test against them and say so.
- Escalate real blockers (missing credentials, environment down, destructive step needed) instead of silently working around or giving up.
- Report coverage honestly: what was tested, what was not, and why. A partial job must not read as complete.
- Respect boundaries from every underlying skill (authorization, no unscoped chaos, no real PII).
Final report format
Verdict: <ship / ship with risks / do not ship> · Confidence: <low | medium | high>
Scope tested / not tested (with reasons)
Findings: by severity, each with repro steps, evidence, impact (link to tickets)
Non-functional notes: performance, accessibility, security observations
Risks and recommended next steps · Suggested regression tests to add
Related
qa-skill-router, qa-fundamentals, exploratory-testing, test-strategist, bug-reporting, release-readiness-testing