Using AgentSecrets with OpenClaw

The agentsecrets skill for OpenClaw (published on ClawHub) is an assistant profile that configures the rules, prompts, and CLI dependencies for the OpenClaw AI assistant. It empowers the assistant to interact with AgentSecrets directly on the host machine.

Because OpenClaw assistants execute local terminal commands, the assistant can autonomously manage your credentials lifecycle. It can inspect configuration status, list key metadata, perform diffs, and sync changes (push/pull) with your cloud backend. The agent never sees the decrypted secret values, upholding the zero-knowledge security model.

Installation

1. Install the skill from ClawHub:

   openclaw skill install agentsecrets

2. Verify that the skill is registered and active:

   openclaw skill list | grep agentsecrets
   # agentsecrets active

Autonomous Assistant Capabilities

Once the skill is active, the assistant understands how to execute AgentSecrets CLI commands on your behalf to manage your workflow. It can run:

• Status queries: Runs agentsecrets status to verify the active project, workspace, and environment context.
• Secret listing & diffs: Runs agentsecrets secrets list and agentsecrets secrets diff to inspect keys (names only) and detect drifted configurations.
• Syncing: Runs agentsecrets secrets pull and agentsecrets secrets push to synchronize local OS Keychain states with the remote sync service.
• Environment switches: Runs agentsecrets environment switch <env> to swap active contexts.
• Secure API calls: Runs agentsecrets call --url <url> --bearer <key> to execute API requests with secure credential injection at the transport boundary.

Sensitive write operations like setting a key (agentsecrets secrets set) require direct user password confirmation and cannot be fully automated without human gating. The assistant will guide you to run the set command in your local terminal.

For non-interactive workflow steps and automated tasks, OpenClaw supports the SecretRef system. When a step configuration references a key via SecretRef, OpenClaw automatically invokes the AgentSecrets exec provider to resolve and inject the key.

Setup Steps

1. Store the credential in the local OS Keychain:

   agentsecrets secrets set STRIPE_KEY=sk_live_...

2. Add domain authorization to your workspace allowlist:

   agentsecrets workspace allowlist add api.stripe.com

3. Reference the credential in your OpenClaw workflow YAML file:

   steps:
     - name: fetch-charges
       skill: standard-http
       inputs:
         url: https://api.stripe.com/v1/charges
         auth:
           secret_ref: STRIPE_KEY

Behind the scenes, when this step executes, OpenClaw runs agentsecrets exec internally. The AgentSecrets binary reads the secret reference, resolves the value from the OS Keychain, performs the injection, and returns the response safely. Plaintext values are never written to any OpenClaw configuration file or workflow log.

If you are running legacy OpenClaw workflows that do not support SecretRef and instead look up keys from environment variables, you can spawn the OpenClaw process using the agentsecrets env wrapper:

agentsecrets env -- openclaw run my-workflow

This resolves the credentials from your local OS Keychain and injects them as standard environment variables directly into the child process at launch time.