ContextLock is a security tool; reports about its own weaknesses are the most valuable contributions it can receive.

Please report vulnerabilities privately via GitHub private vulnerability reporting on this repository (Security tab -> "Report a vulnerability"), or by email to security@mindmodel.ai. Please do not open public issues for exploitable weaknesses.

You can expect an acknowledgment within 72 hours. Coordinated disclosure is appreciated; we will credit reporters unless you prefer otherwise.

• Verified means authentic and unmodified, not safe. A trusted publisher shipping harmful instructions is out of scope by design (SPEC.md section 3).
• The threat model (SPEC.md section 4) states explicitly what ContextLock defends against and what it does not (e.g. a same-privilege attacker with arbitrary code execution outside the host tool). Reports that break a claimed defense (T1-T14) are in scope and serious - especially anything that makes verification pass on modified bytes, bypasses anti-rollback, or silently downgrades a protected file to unverified.
• The enforcement mappings for host tools document known residual gaps (docs/claude-code-surface.md, docs/openclaw-surface.md). Reports that find NEW gaps in those surfaces are welcome.

Pre-1.0: only the latest published release receives fixes.