← ClaudeAtlas

live-schedule-probelisted

Measure whether a server's live tests actually RUN — a cron-triggered workflow whose pytest call selects the `live` marker and whose failure reaches somebody — instead of only being marked and excluded from CI with `-m "not live"`. LIVE_UNSCHEDULED when nothing runs them, LIVE_SCHEDULED_SILENT when a red cron reaches nobody, NO_LIVE_TESTS / CLAIMS_EXTERNAL_COVERAGE / UNVERIFIED when nothing could be measured. Deterministic and read-only; run it, do not reason about it.
malkreide/mcp-continuous-auditor · ★ 0 · AI & Automation · score 71
Install: claude install-skill malkreide/mcp-continuous-auditor
# Live-schedule probe Do the live tests run anywhere, or are they only marked? ```bash python scripts/live_schedule_probe.py --target ../zh-education-mcp python scripts/live_schedule_probe.py --target . --format json python scripts/coverage_run.py --probe live-schedule --manifest manifest.json \ --repos-root ~/portfolio ``` Exit `0` scheduled and visible, `2` **finding**, `3` NOT MEASURED, `4` the checkout moved during the run, `127` the harness could not run. ## Why The doctrine is right: mark the tests that talk to the real upstream `live`, and keep them out of the pull-request run with `-m "not live"`, so a foreign 503 cannot turn an unrelated pull request red. The exclusion then produces the blindness the doctrine was written to prevent. `-m "not live"` is not a place where tests run — it is the absence of one. A test that runs nowhere is documentation, and it rots without anything going red, because nothing runs it. These are also the only tests in the repository that can contradict a wrong assumption about the source: every other test asserts against a fixture written from the same assumption. `meteoswiss-mcp`, 2026-07-30: the first live run in months, three of six tests on the floor, the upstream endpoint retired two days earlier. `zh-education-mcp`, 2026-08-03: four of six datasets read under field names the source had stopped using, eight tools answering every query with an empty list, every unit test green. A live run would have contradicted both. Neither