E20 — Security baseline
| State | Closed |
| Start Date | 2026-10-27 |
| Closed Date | 2026-10-30 |
| Feature | State | Iteration | Stories |
|---|---|---|---|
| F20.1 — Baseline decision | Closed | release-7 | 1 Closed |
| F20.2 — Dependency risk | Closed | release-7 | 3 Closed |
| F20.3 — Source and secret scanning | Closed | release-7 | 3 Closed |
| F20.4 — Release provenance | Closed | release-7 | 3 Closed |
| F20.5 — Disclosure and posture | Closed | release-7 | 4 Closed |
A green Sonar quality gate covers first-party static analysis, security
hotspots, coverage and duplication, and nothing else. It does not cover
vulnerable dependencies (Dependabot watches only github-actions, not
uv.lock), the integrity of the pinned tags consumers install (no SBOM, signing
or attestation, though Actions are already SHA-pinned), secrets in git history,
a licence inventory, or a disclosure process (there is no SECURITY.md). Agent-
written code passes style gates easily, so the baseline has to be explicit and
evidenced per release.
This epic defines that baseline and delivers its controls. The goal is that pykit can state "meets the pykit security baseline" with per-release evidence, not "certified safe". Auth-specific security tests (denied paths, privilege boundaries, token expiry) and the auth threat model are out of scope: they stay in E13, which owns them.
F20.1 comes first: every other feature depends on the controls and thresholds it settles.
Considered and rejected
- Treat the Sonar gate as the baseline. It already runs, but it covers first-party code only; the gaps above are outside what it analyses.
- Claim a certification. The aim is a stated, evidenced baseline per release; "certified safe" is not something these controls can support.