Security Policy

August 10, 2026 · View on GitHub

Supported Versions

Only the latest release on the main branch receives security fixes.

Reporting a Vulnerability

Do not open a public GitHub issue for security vulnerabilities.

Open a GitHub private security advisory to report vulnerabilities confidentially.

Include:

  • Description of the vulnerability
  • Steps to reproduce
  • Potential impact
  • Any suggested fixes (optional)

You will receive an acknowledgement within 72 hours.

Scope

In scope

  • Authentication and authorization bypass in the control plane or gateway
  • Secret leakage via memory documents, skill synthesis output, or logs
  • Remote code execution via tool policy bypass
  • Privilege escalation via the approval gate system

Out of scope

  • Issues requiring physical access to the machine
  • Vulnerabilities in upstream dependencies (report to the dependency maintainer)
  • Issues in CLI backends (Codex, Claude CLI, etc.) — report to their maintainers

Security Design Notes

For the full threat model — secrets encryption, the inbound-webhook attack surface, tool-execution approvals, untrusted-content handling, and the package boundary guarantee — see docs/security/threat-model.md.

Secrets

API keys and secrets are stored in an encrypted keychain, not in config.toml in plaintext. Skills and memory documents are scanned for common secret patterns before storage. See docs/security/secrets-keychain-audit-matrix.md.

Skill synthesis

Synthesized skill drafts are scanned for:

  • API key patterns (sk-..., AKIA...)
  • Password values (password=...)
  • PEM private key headers
  • JWT-like tokens

Drafts containing these patterns are discarded before writing to disk.

Tool policy

The tool policy system (require_approval, deny) enforces approval gates for sensitive operations like bash, write, and edit. Policy is enforced in LemonRouter.Policy before any tool executes.

Skills from untrusted sources

Third-party skill sources require explicit trust approval before installation. Official registry skills are signed and verified. See the trust policy in LemonSkills.TrustPolicy.

Security Contacts

RoleContact
Maintainer@z80

Last reviewed: 2026-03-16