securitylisted
Install: claude install-skill SandRiseStudio/musterd
# Security
Own musterd's application-security evidence without turning investigation into disclosure or authority. A Role is responsibility, not permission to touch infrastructure, production, or another Member's Workspace.
## Start from an attributable baseline
- Read `AGENTS.md`, `docs/design/security.md` when present, and the relevant architecture before scanning.
- Confirm identity, Role, clean Workspace state, current `origin/main` SHA, plugin version, and model.
- Use a dedicated, pre-authorized high-stakes audit Lane; claiming it is authorization to scan its stated scope. If no such Lane exists, obtain Nick's blocking approval before opening or running a whole-repository scan. Keep audit work separate from remediation and public-posture Lanes.
- Threat-model before scanning. Record the SHA, model, plugin version, included surfaces, exclusions, coverage gaps, and validation status. Zero findings is not a security guarantee.
- Use the default model for triage and ordinary diff review. Use `gpt-5.6-sol` with `xhigh` for whole-repository or deep scans and consequential validation.
## Keep findings private
Store manifests, candidates, traces, PoCs, and reports in the Codex Security temporary scan directory, never in tracked files. Read-only fan-out may inspect and write temporary scan artifacts; subagents never edit, claim, build, or commit repository work.
Validate non-destructively with the least evidence needed. Never put exploit details in public issues, branch na