qa-metricslisted
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# QA Metrics
A retrospective runs on evidence or on opinions, and without this skill it runs
on opinions. The goal is a short, honest set of numbers with their reasoning — not
a dashboard, and emphatically not a scoreboard for individuals.
## The numbers come from `qa-metrics`, never from memory
Run it in the workspace first and build everything on its output:
```bash
qa-metrics # readable summary
qa-metrics --json # the same, for a report or a comparison
qa-metrics --since 2026-10-01 # one sprint or release
```
It computes every metric below from files: `05-metrics/events.jsonl`, `bugs/*.md`, the
requirement and scenario JSON under `02-analysis/`, `TRACEABILITY.json` under `03-tests/`,
session sheets under `04-results/*/sessions/`, and the current hashes of artefacts the
pipeline recorded. Each number carries its numerator, denominator and source. A metric with
no data prints `n/a` and what would produce it: report it as missing, never estimate it.
If a number looks wrong, fix the input (a defect file, a missing event), not the output.
`events.jsonl` is written only by `qa-event`, `qa-gate --write` and the `qa-stage-gate`
hook, each from a file (a Playwright report, a gate file, a transcript). Append-only for the
same reason `test_data.yaml` is: a correction is a new line naming the one it replaces. A
metrics file somebody can rewrite is a metrics file nobody can cite. In a workspace that
still has older `05-metrics/*.jsonl` files, run `q