Playwright Agentic

Playwright Agentic

Use this entry skill inside a checkout of playwright-agentic-automation when a task spans framework conventions, product behavior, UI/API automation, or impact analysis. It is a routing guide for an external coding agent; it does not invoke an LLM or make changes automatically.

Start safely

1. Read AGENTS.md for repository-wide ownership and validation rules.
2. Run npm run agent:doctor before runtime work when the environment may be new or stale.
3. For a feature question or impact analysis, read knowledge/index.md, run npm run knowledge:check, and query the narrowest relevant term with npm run knowledge:query -- <term> or npm run knowledge:impact -- <term>.

Route by work type

• UI Page Objects, UI specs, selectors, or navigation: read .agents/skills/pw-ui-pom/SKILL.md.
• API services/specs, fixtures, or authentication: read .agents/skills/pw-api-pom/SKILL.md.
• Scripts, Playwright configuration, linting, reporting, or README tooling claims: read .agents/skills/pw-framework-tooling/SKILL.md.
• Knowledge freshness, source relationships, traceability, or impact: read .agents/skills/codebase-second-brain/SKILL.md.
• A cross-layer failure involving UI, API, data, or Playwright evidence: read .agents/skills/full-stack-incident-analyst/SKILL.md.
• A failed Playwright test, diagnosis-only request, or guarded test repair: read .agents/skills/qa-safe-healing/SKILL.md, then treat qa/failure-taxonomy.json and qa/evidence-schema.json as the policy source of truth.

Use the narrowest relevant skill. When a task spans layers, use the knowledge skill first and then the owning implementation skill.

Evidence boundary

Treat source code, configuration, and executed tests as authoritative. The generated knowledge graph is static evidence, not proof that a path ran. Keep selectors in Page Objects, assertions in tests, and business-meaning changes under human review before they enter active product knowledge.

For a failure, diagnose before modifying. Preserve evidence, classify it, and obey the resulting modification decision. Never weaken assertions, skip/fixme/delete tests, swallow failures, or force actions merely to obtain green output. Run npm run qa:guardrails before accepting a repair.

Demonstration

Run npm run agent:demo for the RBAC impact example. It produces a read-only impact report; inspect the returned evidence before editing or running targeted validation.

Run npm run qa:demo for the QA operating-layer walkthrough. It reports declared coverage gaps, graph-resolved test impact, guardrail status, and benchmark-fixture status without fabricating an agent result.