autonomous intelligence
Secure your AI agent with a transaction-safe MCP layer for guarded, approval-gated local file operations.
Free
Works with the AI Agents you already use
autonomous intelligence
Example session with this skill installed
Set up a safe MCP config for Claude Code in C:\dev\project-alpha and show me the tool definitions.
- Read your context and instructions
- Compiled the autonomous intelligence
I have initialized the workspace environment and generated the configuration for Claude Code. The system is now ready to interface with the trusted Broker for guarded file operations.
AI Agent MCP Configuration
To enable safe file operations, add the following to your claude_desktop_config.json:
{
"mcpServers": {
"autonomous-intelligence": {
"command": "C:\\Users\\Admin\\AppData\\Local\\AutonomousIntelligence\\runtime\\Scripts\\autonomous-intelligence-mcp.exe",
"args": [
"--workspace",
"C:\\dev\\project-alpha"
],
"env": {
"PYTHONPATH": "C:\\Users\\Admin\\AppData\\Local\\AutonomousIntelligence\\runtime\\Lib\\site-packages"
}
}
}
}
Registered Tools
| Tool | Purpose | Security Constraint |
|---|---|---|
autonomous_read_file | Read UTF-8 content | Read-only, returns SHA-256 digest |
autonomous_write_file | Create/Replace files | Requires HMAC approval from Broker |
autonomous_recover_incomplete | Reconcile attempts | Idempotent, no blind retries |
autonomous_get_attempt_status | Check Engine state | Read-only access to SQLite journal |
autonomous_list_recent_operations | Audit history | Non-sensitive operation summaries |
Next steps
- Run
autonomous-intelligence --workspace C:\dev\project-alpha brokerin a separate terminal. - Restart Claude Code to load the new MCP configuration.
- Attempt a read operation to verify the connection between the client and the Engine.
Connects securely to your tools. The creator never sees your data.
About this skill
AI agents often require file system access, but giving a LLM raw OS permissions is a significant security risk. This skill implements a transaction-safe local action layer that separates the untrusted planning agent from a trusted execution engine. It ensures that every file write is gated by a dedicated Broker that requires manual or policy-based approval, preventing unauthorized modifications while maintaining a durable audit trail.
What it does
- Guarded File Operations provides read and write tools that are strictly confined to a specific workspace directory.
- Transaction Safety uses a durable SQLite-backed journaling system to track operation attempts and prevent corruption.
- Approval-Gated Writes forces single-use, expiring HMAC approvals for any modification attempt.
- Crash Recovery reconciles incomplete operations without blind retries, marking unproven states as
UNCERTAINto prevent data loss. - Multi-Client Support generates configuration for Claude Code, Cursor, Gemini CLI, and other MCP-compatible environments.
How it works
- Initialize the Runtime by creating a Python virtual environment and installing the core package.
- Launch the Broker in a persistent terminal to manage the trust boundary and approval queue for the workspace.
- Configure the Client by generating the specific JSON or TOML configuration required for your preferred AI editor or CLI.
- Execute Guarded Tools through the agent to read files, request write approvals, or audit recent operations.
Frameworks & tools
Built for Python 3.11+ on Windows systems. It utilizes the Model Context Protocol (MCP) v2 for communication and SQLite for local state journaling.
Why this beats prompting it yourself
Prompting an agent to "be careful" with files does not prevent hallucinated commands or accidental recursive deletions. This skill enforces a hard architectural boundary where the agent physically cannot execute a write without a verified HMAC token from the Broker.
Use cases
- Granting an agent permission to refactor code within a specific project folder.
- Setting up a crash-safe automated documentation agent that requires human sign-off for every change.
- Providing safe file-reading capabilities to a research agent without risk of system-wide directory traversal.
Known limitations
Currently restricted to Windows due to named-pipe IPC and specific path semantics. Requires an active background Broker process to function.
Windows only (named-pipe IPC). Python 3.11+ required. Requires the Broker process to remain running during agent sessions. Current release focuses on local file operations only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 29 days ago
- Free with an Agensi account
Requires