Anthropic Cybersecurity Skills
An independent, community-built library of more than 800 structured cybersecurity skills for AI agents, each mapped to security frameworks such as MITRE ATT&CK and NIST CSF, for authorized testing, defense and education.
The project's own documentation states it is an independent community project, not affiliated with Anthropic PBC despite its name. It packages 818 skills across 34 security domains as agentskills.io-standard files, so an AI agent pointed at the library can get guidance on tasks such as choosing a memory-forensics tool for a suspicious dump or scoping a cloud security breach across providers. Each skill is mapped only to the frameworks relevant to its subject out of six covered: MITRE ATT&CK and NIST CSF 2.0 apply to nearly all skills, MITRE ATLAS and NIST AI RMF apply to AI-security skills, MITRE D3FEND to defensive countermeasures, and the MITRE Fight Fraud Framework to financially motivated fraud techniques.
The library includes offensive and dual-use techniques, such as red-team command-and-control and phishing simulation, which its documentation states are intended only for authorized penetration testing, security research, defense and education against systems the user owns or has explicit written permission to test. It works with more than 20 AI agent platforms, including Claude Code, GitHub Copilot, Codex CLI, Cursor and Gemini CLI.
It is written in Python and licensed under Apache-2.0.