Repolode
Submit
pegasi-ai

pegasi-ai/reins

Security controls that block, pause, and audit AI agent actions before they run, for Claude Code, OpenClaw, and MCP agents.

View on GitHubWebsite
Stars
393
Forks
52
Open issues
19
Last commit
28 days ago

What is it?

Reins is a security layer for AI agents. It evaluates each agent action against deterministic policies before execution, blocking destructive operations such as bulk deletes and requiring explicit confirmation tokens for catastrophic ones. Approval requests can be routed through a terminal or messaging flow.

The project integrates with Claude Code through PreToolUse and PostToolUse hooks, works with OpenClaw, and supports any MCP-compatible agent. Policies are cached locally so enforcement runs offline, and any unhandled hook error results in the action being blocked.

Reins also scans agent configuration files for OWASP ASI10 issues, tracks security drift over time, and writes an append-only JSONL audit log of every decision, approval, and block.

Key features

  • 01Blocks destructive agent actions before execution and scores their irreversibility.
  • 02Routes high-impact actions to terminal or messaging approval flows.
  • 03Requires explicit CONFIRM-* tokens for catastrophic operations.
  • 04Integrates with Claude Code hooks, OpenClaw, and MCP-compatible agents.
  • 05Keeps an append-only audit trail of decisions, approvals, and blocks.
  • 06Scans agent configs for OWASP ASI10 vulnerabilities and tracks drift.

Use cases

  • Preventing an AI agent from bulk-deleting email or files without approval.
  • Adding human-in-the-loop approval to Claude Code tool calls.
  • Auditing what an autonomous browser or messaging agent actually did.
  • Checking agent configuration files for security weaknesses before deployment.

Quick start

  1. 1

    Install the Reins CLI globally from npm

    npm install -g @pegasi-ai/reins
  2. 2

    Initialize Reins in your project

    reins init

Things to know

  • Requires Node.js 18 or newer.
  • Policy enforcement depends on hook integrations such as Claude Code, OpenClaw, or MCP agents.

In collections

Claude Code tools, skills & pluginsOpen-source AI agent frameworks & appsMCP servers

Topics

#ai-safety#agent-security#human-in-the-loop#audit-trail#mcp#claude-code#ai-agents#agent-observability#ai-monitoring#browser-automation

FAQ

What is reins?
Security controls that block, pause, and audit AI agent actions before they run, for Claude Code, OpenClaw, and MCP agents. Reins is a security layer for AI agents. It evaluates each agent action against deterministic policies before execution, blocking destructive operations such as bulk deletes and requiring explicit confirmation tokens for catastrophic ones. Approval requests can be routed through a terminal or messaging flow.
Is reins free to use? What license does it have?
reins is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is reins still maintained?
Actively maintained. The last commit to reins was on Oct 11, 2026.
What can I use reins for?
Preventing an AI agent from bulk-deleting email or files without approval. Adding human-in-the-loop approval to Claude Code tool calls. Auditing what an autonomous browser or messaging agent actually did. Checking agent configuration files for security weaknesses before deployment.
How do I install reins?
Install the Reins CLI globally from npm: npm install -g @pegasi-ai/reins Initialize Reins in your project: reins init
Where can I get reins?
The source code and installation instructions are on GitHub: https://github.com/pegasi-ai/reins

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects