🪰 Shoofly

AI Agent Runtime Security for Claude Code & OpenClaw

Claude Code is powerful but unprotected. Shoofly adds the safety layer — blocking prompt injection, malware, credential theft, and runaway token spend before they hit. Free to start.

See Plans Stay in the Loop

Quick Start

Works with Claude Code, Cowork, and OpenClaw

Shoofly Basic 🪰🧹 FREE
$ curl -fsSL https://shoofly.dev/install.sh | bash

No account. No credit card. No data sent.

OpenClaw agents now cost real money per token. Don't let a rogue agent run up your bill.

Shoofly Benefits

What Shoofly Blocks

🎯 Blocks Prompt Injection

Stops malicious instructions hidden in web content, emails, and documents from hijacking your agent. Catches 8 injection patterns including jailbreaks, instruction overrides, and base64 payloads.

🔐 Blocks Credential Theft

Intercepts attempts to read or exfiltrate your API keys, GitHub tokens, and AWS credentials before they leave your machine.

🚫 Blocks Unauthorized Writes

Stops writes to sensitive paths — /etc/, ~/.ssh/, LaunchAgents — before your system is modified.

🔁 Stops Runaway Agents

Detects agents stuck in loops: repeated tool calls, call floods, read-write cycles, URL hammering. With pay-per-token billing now the norm for third-party harnesses, loop detection isn't just a security feature — it's cost control.

🔍 Blocks Malware in Tool Results

Catches malicious instructions embedded in web fetches, API responses, and file reads — the attack vector Anthropic calls their #1 risk.

🔑 Finds Secrets Before Attackers Do

One-shot scan of your project directory for 10 credential patterns. CI-integrable.

Why Shoofly

👁️ Full Audit Trail

Every tool call logged to local SQLite. Query your agent history, not just threats.

📖 We show our work

Policy-as-code in YAML. Read it, fork it, audit it. No black box.

⚡ Free Forever

No account. No credit card. No data sent.

🛡️ Stop Threats Before They Fire

Basic detects and alerts. Advanced intercepts before the tool call executes — not after.

📡 Multi-channel alerts

Telegram, WhatsApp, macOS notifications, terminal, OpenClaw gateway. You choose where threats surface.

🏠 100% local

Detection runs on your machine. No cloud API, no DPA, no data retention policy to worry about.

Shoofly Pricing

Runtime security for OpenClaw and Claude Code agents. Every token costs real money now — don't let a runaway agent find out the hard way.

FREE

Shoofly Basic 🪰🧹

Detects threats, sends notifications, never blocks. See exactly what your agents are doing.

  • Works with OpenClaw, Claude Code, Cowork & Dispatch
  • Detects Prompt Injection
  • Detects Credential Theft
  • Detects Unauthorized Writes
  • Detects Runaway Behavior
  • Detects Malware (in tool results and web responses)
  • Multi-channel alerts (Telegram, macOS, terminal)
  • Local SQLite audit trail
  • YAML policy-as-code (open, forkable)
  • Secret detection at rest
  • Unlimited detections, forever
  • No account required

Get started — Shoofly Basic is free:

$ curl -fsSL https://shoofly.dev/install.sh | bash
$5/MO

Shoofly Advanced ⚡🪰⚡

Threats intercepted before they execute — blocked, not detected.

  • Everything in Basic, plus pre-execution blocking
  • Blocks Prompt Injection (pre-execution, not post-detection)
  • Blocks Malware (in tool results and web responses, before your agent acts)
  • Blocks Credential Theft (before the tool call fires)
  • Blocks Unauthorized Writes (before they execute)
  • Where Anthropic's built-in filters stop, Shoofly blocks
  • Zero manual response needed (automatic, 24/7)
  • Custom block/allow policies (YAML, open rules)
  • Policy linting
  • Month to month, cancel anytime
Get Advanced

Stay in the Loop

Get updates on new features, security intel, and the occasional Shoofly wisdom. No spam. Unsubscribe anytime.