← ClaudeAtlas

ccqa-recordlisted

Create a new ccqa test case: pin down the behavior to verify, write its spec.yaml, choose deterministic or live mode, record it if deterministic, and run it until it passes. Use when asked to add, create, or record a new ccqa spec/test for a behavior no existing spec covers.
shibukazu/ccqa · ★ 4 · Testing & QA · score 68
Install: claude install-skill shibukazu/ccqa
# Record a new ccqa test case You are asked to cover one behavior with a new test case. Finish with a spec that passes **for the right reason** — assertions that verify the behavior as the product actually renders it, not whatever happened to be on screen. ## Rules that hold throughout - **Never edit a generated artefact** — `test.spec.ts`, `ir.json`, `generated.json`, anything a target emits. The only file you own is `spec.yaml`; everything else changes by re-recording. - **Follow the neighbors.** Existing specs under `.ccqa/` carry the project's conventions — target, login blocks, session names, naming, serial groups. Read two or three from the same area before writing anything. - **Data the test creates is named with `${CCQA_RUN_ID}`** — ccqa sets a fresh value per spec per run — and a cleanup step at the end of the spec deletes it. A spec that leaves data behind poisons the next run. - **Keep values out of the spec.** URLs, accounts and secrets stay as `${VAR}` references the profile resolves; never inline a concrete value. - **Do not pass `--report-to-hub`** to a local run. Authoring work must not land in the ledger CI reads. ## Before the first command - **How this project invokes ccqa** — `pnpm exec ccqa`, `npx ccqa`, a `package.json` script, or a global `ccqa`. - **Which `--hub-profile`** — the named bucket of variables and sessions for the environment under test. The project's CI answers both: find where it runs `ccqa run` and copy the invoc