dent8 is a memory-integrity firewall for coding agents — a security tool — so we take reports about its own integrity seriously.

dent8 is pre-1.0 (v0.x): the API, the on-disk event encoding (CANON_VERSION), and the storage schema may change between minor versions. Treat it as experimental.

What dent8 does and does not defend against is documented precisely in docs/threat-model.md. In particular, note the stated assumptions:

• The hash chain is tamper-evident; tamper-resistance holds only under an external witness deployment (an anchor issued at write time, key held off the writer's machine). The anchor / verify_signed_head functions are that primitive, not a hosted service.
• The firewall governs provenance, authority, freshness, and contradiction — it does not judge truth, and a compromised high-authority actor is out of scope.

A report that a documented assumption does not hold — or that an undocumented bypass exists (e.g. an un-arbitrated write path, a way to launder authority past arbitrate_events, a canonicalization collision, or a chain/anchor forgery) — is a security issue we want to hear about.

Please do not open a public issue for a security vulnerability.

• Preferred: open a private report via GitHub's "Report a vulnerability" (Security → Advisories) on https://github.com/xyzzylabs/dent8.
• We aim to acknowledge a report within 5 business days and to agree on a disclosure timeline with the reporter.

When you report, please include: the version/commit, a minimal reproduction (ideally a failing test or a sequence of dent8 commands / events), and which threat-model assumption you believe is violated.