Skip to content

Security: denn-gubsky/ai-dev-effectiveness

Security

SECURITY.md

Security

If you discover a vulnerability — for example, a way to leak repository contents through the AI-judge subprocess, or a path-traversal in the config loader — please email the maintainer at project-hockeybot@waverleysoftware.com rather than opening a public issue.

There is no bounty. We aim to acknowledge reports within 7 days and ship a fix within 30.

Threat model notes

  • The default claude-cli judge spawns claude --print as a subprocess in the user's repo directory. The bundled subagent's tools allowlist limits what it can do (Read, Grep, git show/git diff/git log, optional ast-index MCP). It does not include Write, Edit, or unrestricted Bash. If you need to lock this down further, edit .claude/agents/effort-judge.md after init-judge.
  • Diffs sent to opt-in API providers (anthropic-api, openai) leave your machine. ollama keeps everything local.
  • The on-disk cache at .ai-dev-effectiveness-cache/ contains commit SHAs, diffs, and judgments. Add it to .gitignore (the tool does this automatically when you run init-judge).

There aren't any published security advisories