@qa-ai-stlc/runner-playwright

Implements the Runner interface (@qa-ai-stlc/core) for e2e test cases (ADR-004: the engine owns the browser). Given a spec set — absolute paths to hand-written or generated Playwright .spec.ts files — it spawns the real Playwright Test runner directly through its resolved CLI entry point (no npx, no shell), against an ephemeral, import-free config it generates so a spec's own node_modules resolution is never affected, then parses the real JSON report the run produces into validated RunResult values.

A test case id is read from a testCaseId annotation on the spec itself:

test(
  'user can log in',
  { annotation: { type: 'testCaseId', description: 'demo-app-login' } },
  async ({ page }) => {
    // ...
  },
);

A test with no such annotation makes run() throw a QaError — the engine never guesses which case a result belongs to.

run() returns a RunnerOutcome per result — the RunResult paired with whatever raw evidence (a screenshot, a trace) Playwright captured for it. Screenshot-on-failure and trace-on-failure capture are always on; this runner never writes evidence to disk itself (RunResult.evidenceIds is always empty coming out of it) — the caller (qa run / MCP qa.run, via runTestRun) registers the returned evidence through EvidenceStore and fills in evidenceIds (P3-03, ADR-004).

When RunnerInput.identity is set, the runner signs in as that configured identity before the specs start and hands Playwright the resulting session as storageState. The session file lives in the run's temporary directory only until the run ends, traces are turned off for the run (a trace records every request's cookies), and a login that produced no session fails with RUN_SIGN_IN_FAILED instead of running the specs against the login page (P6-89).

Part of QA-AI-STLC. See the repository root for license, contributing and security information.