If you discover a security vulnerability in agentihooks, please report it responsibly.
Do NOT open a public GitHub issue for security vulnerabilities.
Instead, email security@thecloudclockwork.com with:
- Description of the vulnerability
- Steps to reproduce
- Potential impact
- Suggested fix (if any)
We will acknowledge receipt within 48 hours and aim to provide a fix or mitigation within 7 days for critical issues.
agentihooks runs as a hook system and MCP server inside Claude Code agent containers. Keep in mind:
- MCP tools execute with the agent's permissions -- tools like
storage_upload_pathoperate under the agent's credentials - Integration credentials (GitHub App, AWS, SMTP, etc.) are read from environment variables, never hardcoded
- No telemetry or phone-home code exists in this project
- Redis connections are optional and default to localhost
- Never commit
.envfiles or credentials (.gitignorecovers this) - Review
profiles/*/settings.overrides.jsonbefore pushing if you add custom permissions - Use
ALLOWED_TOOLSorMCP_CATEGORIESto restrict tool exposure in sensitive environments - Run the MCP server on localhost only; do not expose it to the network
| Version | Supported |
|---|---|
| 0.1.x | Yes |