artifact-retentionlisted
Install: claude install-skill YevheniiaQA/qa-agent-toolkit
# Artifact Retention
A QA repository grows by tens of megabytes per run and nobody notices until a
clone takes four minutes. The fix is not deleting history — it is separating two
kinds of output that look similar and behave completely differently.
## The two classes
**Disposable run artefacts.** Large, reproducible, only interesting while
investigating a specific failure. Traces, videos, screenshots, raw reporter JSON,
HAR captures, k6 raw output, scanner output, mutation backups. **Retain the last
three runs per target. Never commit them.**
**Durable derived records.** Small, irreplaceable, the basis of every trend and
comparison the pipeline can make. One summary line per run, baselines, scenario
JSON, RTM, the generated report, evidence attached to an open defect. **Keep
indefinitely, commit them.**
The distinction is not about age or size. It is about whether the thing can be
regenerated: a trace can be produced again by re-running; the fact that the p95
was 840 ms on 14 August cannot.
Deleting durable records to save space is a bad trade at any scale — a year of
metrics history is smaller than one Playwright video.
## Retain per target, not per project
A project has several services and interfaces: an API, a landing page, an admin
panel, maybe a second API. "Keep the last three runs" is ambiguous across them,
and in practice a noisy target evicts a quiet one's artefacts — the admin suite
runs twenty times a day, the landing suite twice a week, and last week's la