Skip to content

Repository navigation

Security: tenqaz/agentForge

Security

docs/security.md

Security Checklist

API and Frontend

  • Secrets must not be returned to the frontend.
  • Provider API keys must not appear in JSON responses.
  • WEIXIN_TOKEN and bot_token must not appear in JSON responses.
  • Password hashes must not appear in JSON responses.

Runtime Isolation

  • Each Agent has its own host-side hermes-home.
  • Each Hermes container mounts only that Agent's hermes-home.
  • Runtime container names are generated from internal Agent IDs, not user input.

Weixin Policy

  • Private-message policy is WEIXIN_DM_POLICY=allowlist.
  • Group policy is WEIXIN_GROUP_POLICY=disabled.
  • The confirmed扫码 user ID is written into WEIXIN_ALLOWED_USERS.

Persistence Expectations

  • Deleting and recreating the runtime container must not delete:
    • config.yaml
    • .env
    • sessions/
    • weixin/accounts/{account_id}.json

Admin Template Surface

  • Admins can add complete skills by uploading a ZIP archive and delete complete skills.
  • MVP does not expose a skill edit endpoint.
  • MVP does not expose a skill replace endpoint.

There aren't any published security advisories