Beta version — Coming Soon
Runtime decision layer · Monitor → Approve → Enforce

See Every Risky Action. Block What Hurts.

The runtime security layer for AI agents — monitor, approve, and enforce every action, with skill-poisoning detection built in.

Request Beta Access → ★ View on GitHub
Live Agent Decisions Streaming
0ms
Median decision latency
0+
Malicious skill variants detected
0
Threat signatures & growing
0%
Actions logged & auditable
Live Demo · See It Work

Scan Intent. Block Poison.

Watch ReaperShield scan a skill for a hidden payload and score a live agent action — before anything runs.

agent_guard.py — reapershield Live
Platform Features

Preventive Shield.

Everything you need. Nothing you don't.

🎯

Risk Scoring Engine

5-signal scoring (0–100) on action type, data sensitivity & blast radius. Sub-50ms.

👁️

Monitor-Only Mode

Agents run freely while we detect and alarm. Flip to enforce when you're ready.

🧬

Skill Poisoning Scanner

Scan every SKILL.md, MCP config & Cursor rule before install. DDIPE detection built in.

🔔

Instant Alerts

Telegram, Slack, Email, Webhook — with a one-click "Send Test Alert" on every channel.

📜

Immutable Audit Trail

Every action logged with full context. Export CSV. SOC-2 ready.

🛑

Emergency Kill Switch

One button halts all agent activity instantly, with a forensic snapshot.

Control The
Intelligence Frontier.

🚀 Beta access is opening soon — get on the list.
Free to self-host foreverNo credit cardMIT LicenseUpgrade anytime
Request Beta Access → ★ View on GitHub