Skip to content

Security: Balaji91221/memorylint

SECURITY.md

Security

This is a local development release. Scans are read-only. Index writes, provider commands, and patch application require explicit caller configuration. The MCP surface is read-only and bound to one repository and configured memory paths.

Repository and model text are untrusted data. Paths are checked against repository and claim scope; files and protocol frames are bounded. These checks are not an OS sandbox against a hostile process concurrently replacing directories. Use an immutable Git revision for reproducible inspection, and isolate untrusted provider commands with your platform's process/container controls. Do not expose the stdio server as an unauthenticated network endpoint.

The optional YAML CI parser uses PyYAML's BaseLoader with no object constructors. Explicit tags, aliases/anchors, merge keys, duplicate mappings, and multi-document streams are rejected. Parsing is bounded by the repository's 2 MiB file limit plus 64 nested nodes and 20,000 total nodes. It reads declarations only; workflow steps are never executed. The base installation needs no YAML dependency.

Memory, patch files, and the SQLite index can contain source excerpts. Keep them private, apply repository access controls, and delete the index plus its SQLite sidecar files when removing local history. Provider bridges may transmit selected source text externally; inspect/configure them before use. Credentials should be provided through the chosen provider's environment, not source files or reports.

Reporting a vulnerability

Use GitHub private vulnerability reporting to contact the maintainers. Include the affected version, a sanitized reproduction, and the expected and observed behavior. Do not post secrets, credentials, or sensitive exploit details in public issues.

Security fixes target the latest development release (currently 0.2.0). Older releases have no maintenance guarantee. No independent security audit or response-time guarantee is claimed.

There aren't any published security advisories