Security fixes target the current master branch. Update to the latest reviewed commit before relying on a fix; feature branches are development snapshots.

Use GitHub's private vulnerability report. Include the affected commit, operating system, a minimal reproduction using synthetic data, and the observed impact. Do not post real credentials, private files, exploit targets, or malware samples in public issues.

Run the tools with the least privileges needed. Antivermis reports indicators; it cannot certify a clean host or detect every kernel rootkit. Scanning reads regular files under work limits and does not unpack archives or execute samples.

MCP provides read-only collection under configured filesystem roots. Configure small roots that exclude credentials. Preview redaction is heuristic, not a guarantee that arbitrary secrets will be detected. The tools are not an operating system sandbox: an attacker able to replace parent directories during traversal may race pathname checks. Use an isolated account or read-only snapshot for actively hostile, mutable trees, and do not run the MCP server as root.

Signature manifests rely on the configured HTTPS publisher and their advertised SHA-256; they are not cryptographically signed. Offline file:// updates should use trusted local files. Do not treat a checksum as proof of publisher identity.

GitHub CodeQL default setup scans C/C++, Python, and Actions. Dependabot checks dependencies and action updates, while dependency review rejects newly introduced high/critical advisories in pull requests. Actions are pinned to verified commit IDs and receive only the permissions their job needs. Live settings and scan results are visible in the repository Security and Actions pages; a workflow file alone does not prove a scan has run or passed.