Nullsec Interlock

The execution firewall for AI agents.

Nullsec Interlock sits between AI agents and dangerous actions — shell commands, file writes, git operations, HTTP requests, wallet transactions, database queries, and cloud deploys. Every action is classified, risk-scored, secret-scanned, simulated, policy-checked, and either allowed, blocked, sandboxed, or escalated for human approval — with a full signed audit trail. Everything runs locally.

Quick start

pnpm add -g @nullsec/cli
nullsec init        # scaffold nullsec.policy.ts
nullsec dev         # start local approval daemon
nullsec run -- git push origin main

Features

• Policy engine — glob/wildcard/domain matching rules per action type
• 7 action guards — shell, HTTP, filesystem, git, wallet, database, cloud
• Risk scoring — weighted factor model producing a 0–100 score
• Secret scanning — 12 built-in detectors (AWS, GitHub, Stripe, JWT, PEM, …)
• Simulation — dry-run preview of destructive actions before execution
• Signed audit trail — ES256-signed events with SHA-256 hash chaining
• Local approval daemon — HTTP server for approve/deny from CLI or browser
• MCP server — 7 tools over stdio transport for Claude Desktop / Cursor
• Node.js SDK — @nullsec/interlock for programmatic integration

CLI commands

Command	Description
nullsec init	Scaffold a nullsec.policy.ts in the current project
nullsec dev	Start the local approval daemon
nullsec run -- <cmd>	Execute a command through the interlock
nullsec check <cmd>	Classify and score a command without executing
nullsec approve <id>	Approve a pending action
nullsec deny <id>	Deny a pending action
nullsec verify	Verify audit chain integrity
nullsec timeline	View the audit trail
nullsec policy	Print the resolved policy

MCP setup

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "nullsec": {
      "command": "nullsec",
      "args": ["mcp"]
    }
  }
}

Cursor

Add to your MCP settings:

{
  "nullsec": {
    "command": "nullsec",
    "args": ["mcp"]
  }
}

Packages

Package	Description
@nullsec/interlock	SDK — NullsecInterlock class
@nullsec/cli	CLI binary
@nullsec/core	Shared types, action model, errors
@nullsec/policy	Policy engine
@nullsec/classifier	Action classification
@nullsec/risk	Risk scoring
@nullsec/simulator	Action simulation
@nullsec/secrets	Secret detection
@nullsec/audit	Signed audit events + hash chain
@nullsec/mcp	MCP server (stdio transport)
@nullsec/shell-guard	Shell command guard
@nullsec/http-guard	HTTP request guard
@nullsec/fs-guard	File system guard
@nullsec/git-guard	Git operations guard
@nullsec/wallet-guard	Wallet transaction guard
@nullsec/db-guard	Database query guard
@nullsec/cloud-guard	Cloud deploy guard

Roadmap

See ROADMAP.md for the release plan.

License

MIT