← ClaudeAtlas

verifylisted

Run the full ccdrift suite plus the privacy, determinism and status-line-weight invariant checks, and report only what was actually observed. Use before claiming work is complete, before tagging a release, and after any change to logs.py, history.py, state.py, report.py, draft.py or page.py.
rkolesnichenko/ccdrift · ★ 0 · AI & Automation · score 68
Install: claude install-skill rkolesnichenko/ccdrift
Run each step and paste the real output. Do not summarize a step you did not run, and do not call anything passing without the line that says so. ## 1. Full suite, CI parity ``` uv run --group dev --group lab pytest -q ``` Expect 578 passing at present, roughly 35s. The `lab` group is required even if only `tests/` matters; lab collection imports matplotlib. If the count dropped, find out which test disappeared before doing anything else. ## 2. Status line stays import-light `ccdrift status --short` runs on every status-line refresh, so the status path must import neither pandas nor numpy. pandas alone costs about 0.3s. ``` uv run --group dev --group lab pytest -q tests/test_status.py::test_status_short_loads_neither_pandas_nor_numpy ``` If a heavy import crept in, the fix is to defer it into the command branch in cli.py, not to relax the test. ## 3. Privacy invariants These emit aggregates only: no paths, session ids or project names. `peek` reports text, ids and paths as lengths, and omits the transcript path because it names the project folder. ``` uv run --group dev --group lab pytest -q tests/test_report.py tests/test_draft.py tests/test_page.py tests/test_cli.py ``` ## 4. Determinism Every frame is sorted with `kind="stable"` before any positional or first/last logic, and lab sweeps seed explicitly. If the diff touched a sort, a groupby or a sweep, run the lab gates and confirm the numbers still match docs/findings.md: ``` uv run --group dev --group lab pyt