Skip to content

Security: Oneiros667/CodeMesh

SECURITY.md

Security Policy

CodeMesh indexes repository source and returns selected context to coding agents. Secret handling, repository boundaries, and provider selection are part of its security surface.

Report a vulnerability or exposed secret

Do not open a public issue containing a credential, private source excerpt, exploit detail, customer identifier, or sensitive path. Use GitHub private vulnerability reporting if it is enabled for the repository. Otherwise contact the repository owner through a private channel and include only the minimum information needed to reproduce the issue.

For a suspected committed secret:

  1. Revoke or rotate the credential before treating repository cleanup as remediation.
  2. Identify affected file paths and commits without copying the secret into tickets, chat, or logs.
  3. Preserve evidence privately and coordinate any history rewrite with the owner. Do not force-push a remediation without explicit approval.

There is not yet a published release or supported-version matrix. Security reports should identify the commit tested.

Safe local operation

  • Keep .env, model caches, database volumes, indexes, evaluation traces, and generated source context out of Git.
  • Set a unique local Neo4j password before starting Docker Compose. Default host port bindings are loopback-only; review them before changing that boundary.
  • Use CODEMESH_MODEL_PROVIDER=none and leave summary generation disabled for the source-local baseline.
  • Treat embeddings and generated summaries as copies or derivatives of source content. Confirm repository authorization before enabling any network or third-party model provider.
  • Do not use --keep-raw-traces for shareable evaluation evidence. Raw traces can contain source, prompts, diffs, and tool responses.

Implemented redaction behavior and its known limitations are documented in Security and Redaction.

There aren't any published security advisories