Add ClawGuard — Prompt injection scanner (216 patterns, 15 languages, MIT)

[joergmichno]

ClawGuard Shield

GitHub: https://github.com/joergmichno/clawguard
Website: https://prompttools.co
License: MIT

Overview

Open-source regex-based prompt injection scanner for MCP servers and AI agents.
245 detection patterns, 15 languages, 12 evasion defenses, F1=99.0%.

Sub-10ms scan time, no LLM dependency, fully deterministic.

Would complement your registry

Your registry scores MCP servers on config-level security (tool poisoning, toxic flows).
ClawGuard adds input-level scanning — detecting prompt injection in user/tool inputs before they reach the LLM.

Install

pip install clawguard

Links

• GitHub: https://github.com/joergmichno/clawguard
• API: https://prompttools.co/api/v1/scan
• Blog: https://prompttools.co/blog/prompt-injection-evasion-techniques

[joergmichno]

Quick update: ClawGuard is now at 216 patterns (was 245 — we removed duplicates and improved detection quality). Also added to the description: F1-Score 99.0%, 262 benchmark test cases, OWASP LLM Top 10 + Agentic Security Top 10 fully covered. Happy to update the PR/issue title if needed.