Stop AI agents from doing things you didn't ask for.

Reins watches every action your agent takes - and stops the ones you'd regret. Free and open source.

npm install -g @pegasi/reins

GitHub npm Docs Discord

# 1. Add marketplace & install plugin (skill + hooks)

$/plugin marketplace add pegasi-ai/reins

$/plugin install reins

# 2. Install enforcement layer

$npm install -g @pegasi/reins && reins init

~/projects/revenue-report — reins

Live demo

What it does

Three capabilities. One skill.

Enforce every action

Policy-based enforcement at runtime. Filesystem writes, shell commands, browser actions, and MCP calls evaluated against policy. Destructive patterns blocked, ambiguous ones require human approval.

reins policy

ii.

Scan your environment

13 high-signal security checks. Gateway binding, API key exposure, file permissions, sandbox isolation, sensitive directory access, CVE detection. Auto-fix what's safe. HTML report for your security team.

reins scan

iii.

Monitor for drift

Save a baseline. Diff current state against it. Alerts when posture worsens or config changes without approval. Cron-friendly. Catch permission creep and skill tampering before they become incidents.

reins scan --monitor

Agent-neutral by design

Reins governs any agent that takes actions

Write policy once. Enforce everywhere.

Claude CodeClaude coworkOpenAI CodexCursorHermesOpenClawCustom agents via MCP

Anthropic and OpenAI build the horses. We build the reins.

Mapped to OWASP

Reins maps to 30 risks across three OWASP frameworks for AI agent security. Full coverage matrix documents what's addressed, partial, or on the roadmap.

Full coverage matrix →

MCP Top 10

Security risks for Model Context Protocol servers and tools

7 covered2 partial2 roadmap

Agentic Top 10

Risks in autonomous multi-step agent systems

2 covered5 partial3 roadmap

Skills Top 10

Skill-level risks in agentic runtimes

5 covered5 roadmap

Why Reins?

In Greek myth, Athena gave Bellerophon the golden bridle — reins included — that let him guide Pegasus.

Raw power is not enough. What matters is making it controllable.

The agents are the horses. Reins is the bridle.

For enterprise teams

Reins is the open-source enforcement layer. Pegasi adds the control plane — executive dashboards, framework-mapped evidence for ISO 42001 and AIUC-1, and multi-agent coverage across your environment.