← ClaudeAtlas

qa-reportlisted

Stage 6 — assemble the execution report. All figures come from tooling; the model writes prose only.
alwisofwan1/qa-workflow · ★ 0 · Testing & QA · score 68
Install: claude install-skill alwisofwan1/qa-workflow
# qa-report Input: `results.json`, findings from `qa-heal`, the approved plan. Output: `test-results/<TICKET>-report.md`. ## Division of labour | Produced by | What | |---|---| | `tools/report.ts` | totals, pass/fail/flaky/skipped, AC coverage table, failure list | | you | executive summary, risk assessment, finding narratives, recommendation | Generate the numeric sections by running the tool and pasting its output verbatim: ```bash node dist/tools/report.js results.json --ac AC1,AC2,AC3,AC4,AC5 --require-ac ``` ## Prohibited - **Do not write a number the tool did not produce.** Not a total, not a percentage, not a count of scenarios "planned". - **Do not claim coverage the tool did not compute.** If `report.ts` says an AC is NOT COVERED, the report says NOT COVERED — regardless of how thorough the work felt. - **Do not recommend release when the gate failed.** A non-zero exit from `report.ts` is a blocked release, and the report must lead with that. - Do not describe manual exploratory testing that did not happen. ## Required sections 1. Gate result (pass/blocked) and why — first, before anything else 2. Tool output, verbatim 3. Findings from `qa-heal`, each with assessment 4. Coverage gaps, stated plainly 5. Recommendation, consistent with 1