Skip to content
AgentDock

Last updated October 13, 2026

Security

AgentDock controls agents that write and merge code, so its security matters. If you find a problem, we want to hear about it.

Reporting a vulnerability

Please report security issues privately, not in a public issue.

  1. Open a private advisory

    Go to the Security tab of the repository and choose Report a vulnerability. Only the maintainers can see it.

  2. Include what you can

    The affected component (API, web, runner), the version or commit, steps to reproduce, and the impact you expect. A proof of concept helps, but is not required.

  3. What happens next

    We acknowledge the report, confirm or dispute it, and work on a fix with you. Once a fix is released we publish the advisory and credit you, unless you prefer otherwise.

Scope

In scope: the code in the AgentDock repository, which is the API, the web app, the runner (agentdock-runner) and the shared protocol, plus the reference deployment files.

Out of scope: vulnerabilities in Claude Code, Codex, GitHub or other third-party tools (report those to their vendors), issues that need an already compromised runner machine or admin account, and findings on this marketing website that have no security impact.

Please do not run tests against instances you do not own, and do not access other people's data.

How AgentDock is built to be safe

  • Agent machines open no ports: the runner dials out and only runs a fixed allowlist of typed commands, never shell strings.
  • Model and GitHub credentials never leave the runner machine.
  • Stored secrets are encrypted with AES-256-GCM under a key you generate.
  • Privileged actions land in a hash-chained audit log the database refuses to edit.
  • Signed webhooks in both directions, with replay protection and an SSRF guard on outbound calls.

The full picture, including the known gaps, is in the security model.