agentcheck scans AI agent instruction files for hidden attacks

A developer released agentcheck, a zero-dependency Node CLI and GitHub Action that scans CLAUDE.md, AGENTS.md, Cursor rules and MCP configs for hidden Unicode, injection phrases and exfiltration endpoints. It reported zero false positives across 270 real-world rule files.

• 14 rules cover Unicode Tag smuggling, zero-width chars, bidi overrides and exfiltration
• Caught every attack technique from Pillar Security and AgentSeal research
• Exits 1 on findings and emits SARIF 2.1.0 for GitHub code scanning
• Won't flag plain-language injections without suspicious characters