playwright-e2e-execution-run

Playwright E2E Execution Run

Purpose

This skill executes an existing Playwright end-to-end suite against an operator-confirmed non-production target and emits a structured run attestation: total/passed/failed/flaky counts, slowest tests, retry-only passes, and the location of trace and screenshot artifacts. It is the live-execution counterpart to playwright-e2e-suite-review (which is static-review only and never runs anything). The skill runs the suite as authored — it does not write the tests, deploy the application, or mutate infrastructure — and it refuses to run against a production target.

Execution modes

• Static (default). The skill runs nothing. It inspects playwright.config, enumerates the project and target, states exactly which command it would execute, and asks the operator for explicit runtime opt-in plus target confirmation.
• Runtime (per-session opt-in). Only after the operator explicitly opts in and confirms a non-production base URL does the skill invoke npx playwright test. Runtime mode is never assumed from the request alone.

Lean operating rules

• Never execute the suite without an explicit, in-session runtime opt-in AND an operator-confirmed base URL — absent either, stay in static mode and ask.
• Refuse to run if the target base URL resolves to, or is named like, a production environment (prod, www., a customer-facing apex domain). Require a staging, preview, or ephemeral target; state the refusal reason.
• Never accept credentials, bearer tokens, or a storageState file inline. Test credentials must come from the environment or a config the operator already controls; the skill never collects, echoes, or logs their values.
• Run only the allowlisted commands: npx playwright test (with operator-supplied flags), npx playwright install (browser binaries), npx playwright show-report. Never run deploy, migration, seed, or registry commands.
• Treat the suite's own side effects as the operator's responsibility — state plainly that E2E tests may create or modify data in the target, which is why a non-production target is mandatory.
• Do not retry a failed run with raised timeouts or added retries to manufacture a green result — report the failure as observed.
• Emit the run attestation as JSON conforming to schemas/attestation.schema.json; the verdict is one of pass, fail, or manual-review.
• If browser binaries are missing, run npx playwright install only with operator awareness; if egress to the browser CDN is blocked, degrade to manual-review rather than reporting a false fail.
• Label the run: command executed, target host (host only, never the full credentialed URL), Playwright version, and wall-clock duration.

References

Load these only when needed:

• Workflow and output contract — use when executing the run or formatting the attestation.

Response minimum

Return, at minimum:

• The execution mode used (static or runtime) and why
• The exact command executed (runtime) or that would be executed (static)
• The confirmed target host and Playwright version
• Run results: total / passed / failed / flaky (retry-only pass) counts
• Trace and screenshot artifact locations for any failure
• A pass / fail / manual-review verdict with reasons
• Safe next actions