Submit a Tool

Paste the URL of the tool you'd like to list.

link
Neighbourhood Claw
About Community Documentation API
AgentGuard Security tool screenshot — openclaw.ai

AgentGuard

AgentGuard is an open-source Python security framework for AI agents, built in direct response to the "Clinejection" attack that compromised 4,000 developer machines via a malicious GitHub issue.

Added
1 month ago

About

AgentGuard is an open-source Python security framework for AI agents, built in direct response to the "Clinejection" attack that compromised 4,000 developer machines via a malicious GitHub issue. It detects and blocks prompt injection, command injection (npm install, curl | bash, rm -rf), Unicode bypass attacks (Cyrillic homoglyphs, zero-width characters), social engineering patterns, and encoding/obfuscation techniques — with a surgical sanitization approach that preserves legitimate content while neutralizing threats.

person_check
Best For

Python-based AI agent developers who want layered defense against prompt injection and supply-chain attacks — particularly teams whose agents process external content like GitHub issues, user inputs, or web-scraped text where Clinejection-style attacks are a real risk.

Pros & Cons

check_circle

Pros

  • check Purpose-built for a documented real-world attack vector (Clinejection), not hypothetical threats
  • check Covers an unusually broad threat surface: command injection, prompt injection, Unicode homoglyph attacks, base64/hex obfuscation, and GitHub issue screening
  • check Python library with a clean API — `guard.analyze_text()` and `guard.analyze_github_issue()` return structured threat level and risk score
  • check Surgical sanitization: replaces detected threats while preserving legitimate content rather than blocking entire inputs
  • check Rate limiting and real-time security logging built in
cancel

Cons

  • close Early project (initial commit March 2026, 1 commit total) — not battle-tested in production
  • close Python only — no native support for Node.js/TypeScript agent runtimes
  • close Rule-based detection may have high false positive rates; adversarial inputs evolve faster than pattern lists
  • close Small team (numbergroup) with no documented maintenance commitment or security advisory process

More Security

Other tools in the same category.

View All arrow_forward