QA-CI-008 — Always-success step masks failures
Generated from the rule registry. Get the same detail from the CLI with
mjolnir explain QA-CI-008.
| Field | Value |
|---|---|
| Severity | error |
| Confidence | high |
| Tier | quarantine |
| Measured FP rate | 10% (n=10) |
| Evidence level | E2 |
| QA impact | False-green risk (FALSE-GREEN) |
| False-positive risk (author estimate) | low |
| Autofix available | no |
| Languages | yaml, groovy |
| Frameworks | github-actions, azure-pipelines, jenkins |
| Detection strategy | LEXICAL |
| Introduced in | v0.1.0 |
Why this fails in production
A final step that unconditionally echoes success (or any step that always exits 0 as the LAST step in a job) determines the job's overall conclusion in GitHub Actions — the job's status is the status of its steps taken together, and a trailing always-succeed step can overwrite what an earlier failing step already signaled, depending on how the workflow is structured. Combined with any earlier step's failure being non-fatal to the job (e.g. a missing if: failure() gate), this produces the exact same effect as QA-CI-001 by a different mechanical path: a real failure, a green checkmark.
What gets flagged (real detector output)
Final step in `test` always succeeds while earlier steps tolerate failure.Example from this rule's own must-fire fixture: tests/fixtures/QA-CI-008/must-fire/masked-green.yml
The fix
Remove the always-success final step and let the real test step determine the job result.
Confirmed NOT to fire on the corresponding clean pattern
Verified against tests/fixtures/QA-CI-008/must-not-fire/clean.yml — a legitimate, similar-looking pattern this rule correctly leaves alone.
Corpus-measured false-positive risk
Real occurrence counts from npm run corpus:audit against actively-maintained OSS repos — reproduce yourself, don't just trust this table (see the false-positive audit):
| Repo | Occurrences |
|---|---|
| positive-fixtures | 9 |
| vitest-dev-vitest | 1 |