Security policy

Reporting a vulnerability

If you discover a security vulnerability in secure-code-agent itself (not findings the tool emits about your code — those go in your repo's issue tracker), please do not open a public GitHub issue.

Instead, use the private security advisory flow.

We aim to:

• Acknowledge the report within 72 hours.
• Provide a preliminary assessment within 7 days.
• Coordinate disclosure with the reporter — credit is happily given.

Threat model

See docs/threat-model.md for the full threat model. Short version: the agent runs on developer machines and CI runners; it reads source files, exec's scanner binaries, emits report artifacts. We consider:

• Malicious repo content (no eval / exec / pickle of repo bytes)
• Output injection (markdown / SARIF / PR-comment escaping)
• Scanner output deception (operator-controlled standards mapping)
• Suppression bypass (mandatory reason + expires)
• Baseline tampering (git-attributable diffs)
• Cleartext secret in report (default --redact for secret scanners)
• Scanner network calls (offline modes by default)
• Resource exhaustion (per-scanner wall-clock timeouts)
• Supply-chain attack on this tool (Sigstore-signed releases, pinnable SHAs)

Out of scope

• Runtime attacks against running services (this is a static tool).
• Operator account compromise (out of our control).
• CI runner compromise (gate output is untrusted by definition in that case).
• Compromise of upstream scanners (Bandit, Semgrep, etc.) — we pin versions but do not audit their source.

Supported versions

secure-code-agent is in v0.x pre-release. Only the latest minor version is supported with security fixes during the pre-1.0 phase. Once v1.0 ships, we will publish an explicit support matrix.