Do not open a public issue.
Preferred: Use GitHub's private vulnerability reporting — click "Report a vulnerability" on the Security tab of this repository. This creates a private advisory draft with a CVE workflow.
Alternative: Email [email protected] with details.
Include as much detail as possible:
- Description of the vulnerability
- Steps to reproduce
- Affected component (LLM providers, secret scanning, git operations, config)
- Potential impact
You will receive an acknowledgment within 7 days. Fixes for confirmed vulnerabilities will be released as patch versions with a security advisory.
Security issues in the following areas are in scope:
- LLM streaming — buffer exhaustion, response size limits, malformed server responses
- Secret scanning — pattern bypass, false negatives on known key formats
- Config security — project-level config overrides that could redirect API traffic or exfiltrate data
- Git operations — command injection via file paths, argument injection
- Error messages — credential or URL leakage in error output
- Prompt injection — crafted diff content that manipulates LLM behavior to produce harmful output
- Dependency vulnerabilities — reqwest, tokio, tree-sitter, serde
- LLM output quality (wrong commit type, generic subjects) — use the issue tracker
- Feature requests — use the issue tracker
- Vulnerabilities in Ollama, OpenAI, or Anthropic services themselves