Email security@firstops.dev (or open a private security advisory on GitHub). Please do not open a public issue for security reports. We aim to acknowledge within 3 business days.

whittle rewrites the tool output an AI agent reads. Beyond conventional vulnerabilities, we treat these as high severity:

• Fidelity violations: any input where whittle changes the meaning of an output a lossless path claims to preserve, or where code/config reaches the lossy prose model. (See GUARANTEES.md.)
• Retrieval integrity: whittle_get returning content that is not the exact original for a given id.
• Fail-open breaches: any path where a whittle failure breaks a tool call instead of passing the original through.

All three are pinned by tests; a reproducer that defeats them is a security bug.