← ClaudeAtlas

qa-conventionslisted

Conventions every QA skill shares — where run output goes and how runs are recorded, how a disabled integration is handled, how a lesson is recorded. Read it whenever a QA skill points to it (sections "Where run output goes", "Disabled integrations"), and before writing run artefacts, metric events or anything that would call an outside system from a QA workspace.
YevheniiaQA/qa-agent-toolkit · ★ 0 · Testing & QA · score 72
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# QA conventions Skills point here instead of repeating these rules, so they cannot drift apart. A skill's own section adds only what is specific to it: the command it records with, the systems it touches, its local alternative. ## Run output Write heavy artefacts — traces, videos, screenshots, raw reporter output, scanner and capture files — to `.runs/<target>/<timestamp>/`, where `<target>` is the service or UI target from the project profile. Never to the repository root, never into the generated test project, and never anywhere that gets committed. Namespacing by target is what makes retention work on a multi-service project: without it, a suite that runs twenty times a day evicts the artefacts of one that runs twice a week. Record every run with the `qa-event` command the skill names; it is the only way into `05-metrics/events.jsonl`. Never write that file by hand or from memory: `qa-event` reads the numbers from the file it is given, names that file in the event and validates the line. The file is append-only and kept indefinitely; it costs a few hundred bytes a run and is the basis of every trend `qa-metrics` can show. Only the last three runs per target are retained. If a run produced evidence for a defect that is still open, or is the current baseline, or is the last known-good run, add its path to `.runs/KEEP` so retention does not remove it. See `artifact-retention`. ## Disabled integrations Read `INTEGRATIONS` from the project profile (`source ~/.qa-agent/