Report security issues privately through GitHub's security advisory form, not a public issue. Include reproduction steps, affected versions, and any relevant logs with secrets redacted.

• Token handling: diffly reads GITHUB_TOKEN from the environment or the Action context. Tokens are never logged, echoed, or included in reports.
• LLM boundary: deterministic mode sends no code anywhere. With --explain, bounded, secret-redacted context is sent to the configured OpenAI-compatible endpoint. Model output is strictly validated and cannot change verdicts. See docs/phase-2-contract.md.
• Untrusted input: PR titles, bodies, diffs, and check names are treated as data, never instructions.

Please review docs/phase-2-contract.md before enabling --explain for sensitive repositories.