Skip to content

For coding agents

Teach your agent to request access.

Let your coding agent work with credentials stored on your phone. It requests access for the command it needs; you choose what to allow.

Start with the skill

Give your agent the skill.

The Agentknock skill teaches your agent to install the CLI, pair with your phone, and use secrets for the work you ask it to do. It can also help move existing credentials to your phone.

Any LLM can read the skill; using Agentknock requires an environment that can run the CLI. The plugin packages the same skill for OpenAI and Anthropic marketplaces.

Ask your agent to get started
Read the Agentknock skill at:
https://raw.githubusercontent.com/agentknock/agentknock-plugin/master/plugins/agentknock/skills/agentknock/SKILL.md

Help me set up Agentknock and pair this environment with my phone.

Install the Android app on your phone. Your agent can handle the CLI setup and guide you through pairing. Compare the verification code it shows with your phone, and approve only if they match.

Once the skill is installed, just ask: “Set up Agentknock.”

From your request to a command

Ask for the work. Keep control of access.

“Use Agentknock to review the open issues before we prepare a release.”
  1. Discover available secrets

    The agent can find the right secret by name and description, without reading its value. Use an established name, or list the available secrets and their public metadata.

    Discover names and metadata
    agentknock secret list
  2. Request the command that needs access

    The agent wraps the tool invocation with Agentknock and gives a specific reason for access. Your phone follows the secret’s approval settings, asking you when a decision is needed.

    The command your agent runs
    agentknock -s github \
      --reason "Review issues for the next release" \
      -- gh issue list

    Assumes gh is installed, the working directory is a GitHub checkout, and github contains a sensitive GH_TOKEN.

  3. Wait, then use the result

    Keep the original CLI session while it waits. Read the command’s output and exit status to establish success. On a denial, report it and stop that attempt; do not bypass it or send duplicate requests.

    Approved environment values go to the command, whose code can still read or disclose them. SSH authentication and Git signing keep the private key on your phone.

For agents reading this page

Start with the skill. Follow the task.

Read the Agentknock skill and consult its linked references for installation, pairing or credential migration as needed. You can read those instructions directly without a skill loader or plugin installation.

Work with the user’s setup.

  • Reuse an existing CLI installation and pairing when available.
  • If setup is needed, follow the skill’s installation and pairing guidance. The user verifies and approves pairing on their phone.
  • Use your environment’s permission mechanism when a sandbox blocks network access, pairing storage or local sockets.
  • Run the intended command with Agentknock. Do not print credentials or put their values in the conversation.