A security-first proxy that enables AI agents to access third-party APIs and Git repositories through human-in-the-loop Telegram approvals.
The fastest way to install a skill directly from the registry.
npx clawhub@latest install permissions-broker
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
Copy this prompt to OpenClaw to install it automatically.
Help me install permissions-broker using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Get the raw skill files in a ZIP archive.
The Permissions Broker serves as the default mechanism for external data access within Openclaw Skills when local credentials are unavailable. It functions as an approval-gated proxy, allowing AI agents to propose specific upstream requests (URL, method, and body) which the user must then authorize via a Telegram interface. This architecture ensures that sensitive actions remain under direct human control while providing agents with the connectivity they need.
By leveraging this skill, Openclaw Skills can interact with major providers including Google (Drive, Docs, Sheets), GitHub, iCloud (CalDAV), and Spotify without the agent ever handling long-term user credentials. The broker manages the complex OAuth flows and credential injection, presenting a unified and secure API for the agent to consume data and perform actions across the web.
To integrate this skill with your Openclaw Skills environment, follow these steps:
/key <your_key_name>
/connect <provider_name>
export PB_API_KEY='your_secret_key_here'
The Permissions Broker organizes its operations around request objects and session metadata. The following table outlines the core data structure for proxy requests:
| Attribute | Description |
|---|---|
upstream_url |
The destination HTTPS URL for the external service. |
method |
The HTTP verb used (GET, POST, PUT, PATCH, DELETE). |
consent_hint |
A plain-language string explaining the request to the user. |
status |
The lifecycle state: PENDING_APPROVAL, APPROVED, EXECUTED, or DENIED. |
request_id |
A unique identifier used for polling and execution. |
Constraints: The broker enforces a 256 KiB cap on request bodies and a 1 MiB cap on upstream responses to ensure performance and security.
Loading
A bridge for AI agents to generate shareable links that allow users to securely execute blockchain transactions or sign messages using their own wallets.

Integrate Linear project management into your agentic workflow to automate issue tracking, notification handling, and team synchronization.

A command-line interface for the AgentMail API to manage disposable inboxes and automate email communication.

A competitive gaming platform where AI agents test their logic and strategy through Minesweeper, Checkers, and commit-reveal duels.

A lightweight, zero-dependency Node.js HTTP client for making resilient API requests with automatic retries and graceful error handling.

A sophisticated academic analysis tool that generates both deep technical reports and concise logical summaries for any scientific paper.
