gate-ponytail

# Gate 2: ponytail > **Does anything else in the run already record this?** Named for the [ponytail](https://github.com/DietrichGebert/ponytail) review style: find what to delete. Most of what a spec accumulates is machinery reporting something the tooling already reports. This gate is scoped to over-engineering and duplication only. Correctness, security and performance belong to a different pass. ## The repo facts this gate leans on Check `playwright.config.ts` before citing these, but at time of writing: - **`trace: 'on'` and `video: 'on'` for every test.** The trace already records every request, its timing, and its body. So an attachment or a `test.step` that exists only to surface that information is duplicate. - **`CustomReporter` already prints** per-test status, step timings, the flaky diff, and the AI tabs. A `console.log` restating any of it is noise. - **`createAgent` already validates** its output against a schema. Re-validating downstream is belt-and-braces on a belt. ## Findings format One line each: `<file>:L<line>: <tag> <what>. <replacement>.` | Tag | Means | |:----|:------| | `delete:` | Dead code, unused flexibility, speculative feature. Replacement: nothing. | | `stdlib:` | Hand-rolled thing the standard library ships. Name the function. | | `native:` | Code doing what the platform or framework already does. Name the feature. | | `yagni:` | Abstraction with one implementation, config nobody sets, layer with one caller. | | `shrink:` | Same logic, fewer lines. Show the shorter form. | End with `net: -<N> lines possible.` If there is nothing to cut, say `Lean already. Ship.` ## Worked example from this repo ``` L1-18: delete: 17-line header explaining the rewrite. The commit message holds it. L23-31: delete: testInfo.attach of the response. trace:'on' already records it. L26,36: native: test.step wrappers around single calls. The trace shows each request with timings. L33-57: yagni: try/finally cleaning one row from a public demo API. net: -38 lines possible. ``` ## What this gate must never cut - **Any assertion.** A runnable check is the minimum, not bloat. - **Knowledge that cannot be re-derived from the code.** A comment recording that restful-booker answers a bad token with 403 and bad credentials with 200 took work to obtain. Keep it. - **Test granularity.** How many tests a flow is split into is a decision about how failures report, not machinery. `describe.serial` giving each stage its own pass or fail is a choice. Flagging it as duplication is this gate exceeding its scope. - **File header comments**, which are the convention on every util in `src/utils/`. ## Verdict `FAIL` when the diff adds machinery that duplicates the trace, the reporter or the factory. Report the net line count either way.