Please do not open a public GitHub issue. Report vulnerabilities privately through GitHub's Security Advisory flow:

→ Open a private advisory (github.com/DailybotHQ/ai-diff-reviewer/security/advisories/new)

The advisory is scoped to a small circle of maintainers and lets us collaborate on the fix privately before a coordinated disclosure. You do not need special permissions to open one — any GitHub account can submit.

We aim to acknowledge within 48 hours and ship a fix or workaround within 14 days for high-severity issues. Lower-severity issues are triaged on the same schedule but may take longer to resolve.

We publish releases as SemVer git tags (vX.Y.Z) and maintain a moving major-version alias for the current line (v2). Security fixes ship as patch releases against the current major only; older majors are unsupported.

For the complete trust model, supply-chain notes, secrets handling, per-provider egress surfaces, and known accepted risks, see docs/SECURITY.md.

Highlights covered there:

• Runtime trust boundary (composite action runs in the consumer's runner with the consumer's tokens).
• Per-provider outbound network surfaces (Anthropic API vs the agent-runner CLIs — claude-code, cursor, codex).
• Agent-runner residual exfiltration surface (GITHUB_TOKEN persisted by actions/checkout, vendor API key in the CLI subprocess env) — recommended hardening: persist-credentials: false + trusted/non-fork PRs only.
• author-association gate — public-repo abuse defense, default write-tier (OWNER,MEMBER,COLLABORATOR), evaluated before any LLM API call.
• Log redaction (redact_for_log + LOG_REDACT_SUBSTRINGS) and safe path resolution (safe_repo_path).
• Cursor installer supply-chain note and MCP config persistence caveats.