domains / / awslabs/aidlc-workflows
Security Trust Boundaries
When software ingests extensions (plugins) or external artifacts (actions/images/deps), security standards must explicitly state the trust model and maintenance cadence.
When software ingests extensions (plugins) or external artifacts (actions/images/deps), security standards must explicitly state the trust model and maintenance cadence.
Apply this rule:
1) Document trust boundaries where install equals code execution. If an extension can be composed/loaded from a location without a trust gate (e.g., a “folder-drop” plugin), add a clear security note and user guidance (treat as git clone && run; only accept from sources you would run; review the diff; pin to a reviewed tag/release).
2) Pin supply-chain artifacts, but also define update cadence. Keep SHA/version pinning for reproducibility and supply-chain safety, and set a cadence per component. Scanner/signature services that intentionally freeze data behind a digest must be updated more frequently (e.g., monthly for ClamAV signature DB digests).
3) Require verification after updates. After any pinned update, run the relevant security workflow(s) and confirm outputs (e.g., SARIF uploads/artifacts) and that new findings are expected.
Example language for plugin docs (Kiro folder-drop):
### Security Considerations (Kiro folder-drop)
Kiro plugins are composed on your machine with your shell’s privileges. There is no install-time trust gate—dropping the plugin tree is the trust decision.
Treat a plugin drop like `git clone && run`: only use plugins from sources you would run code from, review the diff, and pin to a reviewed tag/release.
Example cadence rule for pinned scanners:
- GitHub Actions / scanner tools: review at least quarterly.
- Signature-backed scanner images (e.g., ClamAV digest): review at least monthly.
- After updating: run the security workflow and validate outputs.