rjmurillo PR Quality Gate: QA
Run the qa axis of the quality gate against your working changes, before you push, using the same criteria CI will apply.
Migrated from the pr-quality/qa command under ADR-064, which makes skills the single user-invocable surface and renames the namespaced sub-command pr-quality/qa to pr-quality-qa. The command file is gone, so its path is named here in plain text rather than as a citation to something a reader could open.
Triggers
qa review my changes, run the qa gate, are these tests good enough, pr-quality qa
Arguments
Base branch: $ARGUMENTS
If $ARGUMENTS names a branch, diff against it. Otherwise default to main.
Criteria
Apply the criteria from: @.github/prompts/pr-quality-gate-qa.md
That file is the canonical axis definition and the same one CI loads. Do not paraphrase it from memory: read it, then judge the diff against what it says.
Process
- Run
git branch --show-currentto name the current branch. - Resolve the base branch from
$ARGUMENTS, defaulting tomain. - Run
git diff "<base_branch>" --name-onlyto list the changed files. - Run
git diff "<base_branch>"to read the full diff. - Judge the diff against every criterion in the axis file. Findings are additive: a criterion that fails does not end the pass, because the author needs the whole list in one round rather than one finding per round.
- Emit the verdict block, then the JSON block.
What this axis looks for
This axis reads the diff for a new branch with no test, an error path asserted only on a return value, a happy-path-only case, a fixture supplying the precondition the code should establish, and an assertion no input can falsify.
That list orients the read. The axis file is authoritative when the two differ.
Output Format
Provide your verdict in EXACTLY this format. Do not add preambles, explanations, or additional text before the verdict:
VERDICT: [PASS|WARN|CRITICAL_FAIL]
MESSAGE: [One sentence summary]
[Detailed findings following prompt structure]
Then emit a fenced JSON block conforming to .agents/schemas/pr-quality-gate-output.schema.json with "agent": "qa".
Verification
- The axis file was read this run, not recalled
- The diff was taken against the resolved base branch, not against
HEAD~1 - Every criterion in the axis file was judged, including the ones that passed
- Each finding names a file and a line, not a general concern
- The verdict block is the first thing emitted, with no preamble
- The JSON block validates against the schema and carries
"agent": "qa"
Anti-Patterns
| Avoid | Why | Instead |
|---|---|---|
| Stopping at the first CRITICAL_FAIL | Findings are additive, so an early stop hides the rest and buys the author a second round | Record the verdict and keep judging the remaining criteria |
| Judging from memory of the axis | The criteria change and the recalled version is the one that was true once | Read @.github/prompts/pr-quality-gate-qa.md every run |
| A finding with no file and line | The author cannot act on it, so it reads as an opinion and gets skipped | Cite path:line and quote the offending span |
| A preamble before the verdict | The block is parsed by pr-quality-all and by CI, so leading prose breaks the merge |
Emit VERDICT: first, every time |
| PASS on a diff you could not read | An unread diff is an unmeasured one, and PASS says the opposite | Emit CRITICAL_FAIL or UNKNOWN and name what blocked the read |
Extension Points
- New criterion. Add it to
.github/prompts/pr-quality-gate-qa.md, which CI and this skill both read, so the two cannot drift. - Different base. Step 2 resolves the base from
$ARGUMENTS; a project that merges into something other thanmainchanges that default alone. - Another consumer. The JSON block is schema-bound, so a new reader parses it without this skill changing.