Skip to content
GitHub

Credential Brokers

Your agent can use skills that call signed-in services while your tokens stay out of the repository and out of the chat.

You need agentbundle first. Install it once for your user account:

agentbundle install --pack credential-brokers

In Claude Code you can install it as a plugin instead:

claude plugin install credential-brokers@agent-ready-repo
  • Run credentialed skills, such as Jira or Figma, with tokens it never sees.
  • Walk you through storing each key a skill needs, one prompt at a time.
  • Keep each secret in your OS keychain, or in a private file on Linux.
  • Reuse a corporate SSO session for skills that sign in through a browser.
  • Let you add or rotate a key at any time.
  • credential-setup — An interactive setup that prompts you for each key a credentialed skill declares and stores it in the safest place available. You start it yourself.

You also get two helpers that skills call, not you: credbroker, a Python library that resolves credentials in-process, and sso-broker, which handles SSO cookie sessions.

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and Kiro IDE. Atlassian, Figma, and Linear require it for their tokens. To build your own credentialed skill, see Add credentials to a skill.

How do I store a token? Install a credentialed pack, then say “set up credentials”. The skill prompts you, and you type each value.

Does the agent ever see my token? No. Setup runs for you, not the agent, and the value is not echoed into the transcript. The setup script refuses --token, --pat, and --password flags, so a secret never goes on the command line.

Where is it stored? The OS keychain on macOS and Windows. On Linux it is a 0600 file at ~/.agentbundle/credentials.env. Skills also accept an environment variable first.

Do I call the broker myself? No. A credentialed skill resolves its token through it once setup has stored it.

Scope: user (default), repo · Version: 0.3.4 · Source: packs/credential-brokers