There’s a well-documented problem at the heart of AI-assisted security work. General-purpose language models are impressively broad — they know what Kerberoasting is, they’ve read the Volatility3 documentation, they’re aware that SSRF exists — but they lack the structured, step-by-step practitioner knowledge that turns awareness into action. Ask one to perform a memory forensics investigation and it might tell you to “run Volatility3 and look for suspicious processes.” Ask a senior DFIR analyst and they’ll tell you exactly which plugins to run, in which order, what to look for in the output, how to correlate findings across artefacts, and how to verify the results before writing them up.
That gap is what Anthropic Cybersecurity Skills sets out to close. It’s currently the largest open-source cybersecurity skills library for AI agents: 754 structured skills across 26 security domains, mapped to five industry frameworks, compatible with Claude Code, GitHub Copilot, Cursor, Gemini CLI, and more than 20 other platforms. With nearly 9,600 GitHub stars and 1,200 forks, it has clearly found a large and engaged audience.
One important note before we go further: despite the name, this is an independent community project by mukul975 (Mahipal Jangra). It is not affiliated with Anthropic PBC. The name reflects its primary intended platform — Claude Code — rather than any official relationship.
What Is a “Skill” in This Context?
The word “skill” has a specific meaning here, defined by the agentskills.io open standard. A skill is a structured Markdown file with a YAML frontmatter header that an AI agent can discover, load, and execute. It’s not a script, a wordlist, or a blog post summary. It’s an AI-native knowledge unit that encodes a practitioner workflow in a format optimised for agent consumption.
Each skill in the library follows a consistent directory structure:
skills/performing-memory-forensics-with-volatility3/
├── SKILL.md ← Skill definition (YAML frontmatter + Markdown body)
├── references/
│ ├── standards.md ← MITRE ATT&CK, ATLAS, D3FEND, NIST mappings
│ └── workflows.md ← Deep technical procedure reference
├── scripts/
│ └── process.py ← Working helper scripts
└── assets/
└── template.md ← Filled-in checklists and report templatesThe YAML frontmatter is designed for speed. An agent can scan all 754 skill headers at roughly 30 tokens per skill — fast enough to search the entire library in a single pass without consuming its context window. When the agent identifies relevant skills, it loads those in full (500–2,000 tokens each) and follows the structured workflow. This progressive disclosure architecture is a genuinely clever design decision that makes the library practical at scale.
Here’s what a real frontmatter looks like:
---
name: performing-memory-forensics-with-volatility3
description: >-
Analyze memory dumps to extract running processes, network connections,
injected code, and malware artifacts using the Volatility3 framework.
domain: cybersecurity
subdomain: digital-forensics
tags: [forensics, memory-analysis, volatility3, incident-response, dfir]
atlas_techniques: [AML.T0047]
d3fend_techniques: [D3-MA, D3-PSMD]
nist_ai_rmf: [MEASURE-2.6]
nist_csf: [DE.CM-01, RS.AN-03]
version: "1.2"
author: mukul975
license: Apache-2.0
---Below the frontmatter, each skill’s Markdown body covers four sections: When to Use (trigger conditions for the agent), Prerequisites (required tools, access, and environment), Workflow (step-by-step execution with specific commands and decision points), and Verification (how to confirm the skill was executed correctly).
Five Frameworks, One Library
What sets this library apart from other security tool collections is the framework mapping. Every skill is mapped to all five of the major frameworks used in professional security work:
- MITRE ATT&CK v18 — 14 tactics, 200+ techniques covering adversary behaviours and TTPs
- NIST CSF 2.0 — 6 functions, 22 categories covering organisational security posture
- MITRE ATLAS v5.4 — 16 tactics, 84 techniques covering AI/ML adversarial threats
- MITRE D3FEND v1.3 — 267 defensive techniques across 7 tactical categories
- NIST AI RMF 1.0 — 4 functions, 72 subcategories for AI risk management
To illustrate what this means in practice, a single skill like analyzing-network-traffic-of-malware maps simultaneously to ATT&CK T1071 (Application Layer Protocol), NIST CSF DE.CM (Continuous Monitoring), ATLAS AML.T0047, D3FEND D3-NTA (Network Traffic Analysis), and NIST AI RMF MEASURE-2.6. That cross-referencing is something you’d normally have to do manually when writing up a finding or building a detection.
The ATLAS and AI RMF mappings deserve particular attention. MITRE ATLAS covers adversarial threats specifically targeting AI and ML systems — including, in the v5.4 release, agentic AI attack vectors added in late 2025: AI agent context poisoning, tool invocation abuse, MCP server compromises, and malicious agent deployment. As AI agents become more prevalent in security operations, these threat categories are moving from theoretical to immediately relevant. Having them mapped in a skills library built for AI agents is a forward-looking design choice.
The 26 Security Domains
The 754 skills span a genuinely comprehensive range of security disciplines. The largest domains by skill count are:
- Cloud Security (60 skills): AWS, Azure, and GCP hardening, CSPM, cloud forensics
- Threat Hunting (55 skills): Hypothesis-driven hunts, living-off-the-land detection, behavioural analytics
- Threat Intelligence (50 skills): STIX/TAXII, MISP, feed integration, actor profiling
- Web Application Security (42 skills): OWASP Top 10, SQLi, XSS, SSRF, deserialization
- Network Security (40 skills): IDS/IPS, firewall rules, VLAN segmentation, traffic analysis
- Malware Analysis (39 skills): Static and dynamic analysis, reverse engineering, sandboxing
- Digital Forensics (37 skills): Disk imaging, memory forensics, timeline reconstruction
- Security Operations (36 skills): SIEM correlation, log analysis, alert triage
- Identity and Access Management (35 skills): IAM policies, PAM, zero trust identity, Okta, SailPoint
- SOC Operations (33 skills): Playbooks, escalation workflows, metrics, tabletop exercises
Specialist areas are also well-covered: Container Security (30 skills including Kubernetes RBAC, image scanning, and Falco), OT/ICS Security (28 skills covering Modbus, DNP3, IEC 62443, and SCADA defence), API Security (28 skills covering GraphQL, REST, OWASP API Top 10), Red Teaming (24 skills for full-scope engagements, AD attacks, and phishing simulation), and DevSecOps (17 skills for CI/CD security, code signing, and Terraform auditing).
The thinner domains — Deception Technology (2 skills) and Compliance & Governance (5 skills) — are highlighted as priority areas for community contribution if you’re looking for a way to get involved.
How Agents Actually Use the Skills
The library’s progressive disclosure architecture is worth understanding before you set it up. When you give an AI agent access to the skills library and ask it to investigate something — say, “analyze this memory dump for signs of credential theft” — the agent’s internal process works in stages:
- It scans all 754 skill frontmatters (roughly 30 tokens each) and identifies the 10–15 most relevant skills by matching tags, domain, and description.
- It loads the top two or three matches in full — perhaps
performing-memory-forensics-with-volatility3,hunting-for-credential-dumping-lsass, andanalyzing-windows-event-logs-for-credential-access. - It follows the structured Workflow section step-by-step — running specific Volatility3 plugins, checking LSASS access patterns, correlating with event log evidence.
- It validates results using the Verification section, confirming IOCs and mapping findings to ATT&CK T1003 (Credential Dumping).
Without the skills library, the agent guesses at tool syntax and likely misses critical steps. With it, the agent follows the same practitioner playbook a senior DFIR analyst would use — consistently and without the variation you’d get if you just asked the base model to improvise.
Getting Started: Installation
Installation is quick and requires no configuration beyond having a compatible AI platform set up. There are two methods:
Option 1: npx (Recommended)
npx skills add mukul975/Anthropic-Cybersecurity-SkillsThis uses the agentskills.io CLI to install the library and register it with compatible platforms automatically.
Option 2: Git Clone
git clone https://github.com/mukul975/Anthropic-Cybersecurity-Skills.git
cd Anthropic-Cybersecurity-SkillsOnce cloned, point your AI platform at the skills/ directory. For Claude Code specifically, you can place the skills in your ~/.claude/ directory or reference them from a project-level configuration.
Using with Claude Code
Claude Code picks up skills automatically from configured skills directories. Once installed, you can start using the library immediately by describing security tasks naturally in your Claude Code session:
# Threat hunting
"Hunt for signs of lateral movement in these Windows Security logs"
# Memory forensics
"Analyze this memory dump for credential theft activity"
# Web application testing (on authorised targets only)
"Test this login endpoint for SQL injection vulnerabilities"
# Threat intelligence
"Profile this C2 infrastructure based on these network IOCs"
# Cloud security
"Review this AWS IAM policy for privilege escalation paths"
# Incident response
"We have a potential ransomware incident — walk me through the
immediate containment steps"Claude will scan the skill library, identify the most relevant skills for your task, load their workflows, and follow them step by step. You’ll see the structured methodology being applied rather than improvised.
Using with Other Platforms
The library works with any platform supporting the agentskills.io standard. For GitHub Copilot, Cursor, Windsurf, and similar tools, consult their respective documentation for how to register a skills directory. For LangChain, CrewAI, AutoGen, and other agent frameworks, skills can be loaded as context documents and retrieved via semantic search. The index.json file in the repository root provides a structured index of all 754 skills for programmatic access.
MITRE ATT&CK Coverage in Depth
For security professionals who work with ATT&CK regularly, the coverage breakdown across all 14 enterprise tactics is worth reviewing. The library’s strongest coverage is in Execution, Persistence, Privilege Escalation, Defense Evasion, Credential Access, Lateral Movement, Command and Control, Exfiltration, and Impact — the tactics that matter most in real-world engagements. Reconnaissance, Discovery, and Collection are marked as Moderate, reflecting areas where the community can contribute additional skills to improve depth.
An ATT&CK Navigator layer file is included in the v1.0.0 release assets for visual coverage mapping — useful if you want to quickly identify gaps before an engagement or training exercise.
NIST CSF 2.0 Alignment
The library covers all six CSF 2.0 functions, with the heaviest concentration in Detect (200+ skills) and Protect (150+), followed by Respond (160+) and Identify (120+). The Govern function — added in the CSF 2.0 revision of February 2024, expanding scope from critical infrastructure to all organisations — has 30+ skills covering risk strategy, policy frameworks, and roles and responsibilities. Recover has 40+ skills covering ransomware recovery, business continuity planning, and disaster recovery.
Who Is This For?
For security practitioners using Claude Code or other AI coding assistants in their daily work, this library is the missing piece that turns a capable general-purpose model into a consistent, methodical security analyst. The difference between an agent improvising a memory forensics investigation and one following a structured Volatility3 workflow is substantial in both accuracy and completeness.
For students and those preparing for certifications like OSCP, GCIH, GREM, or cloud security credentials, the skills library is an excellent structured study resource. Each skill encodes the kind of step-by-step methodology that certification exams and real engagements both require — and working through skills with an AI agent gives you immediate feedback on whether you understand each step.
For security teams building AI-assisted workflows into their SOC or IR processes, the library provides a foundation that’s grounded in established frameworks rather than LLM improvisation. The D3FEND mappings are particularly useful here: when an agent detects a threat, it can recommend specific countermeasures by their D3FEND identifier rather than giving generic advice.
For researchers and developers building security-focused AI agents, the library provides a production-grade starting point. The agentskills.io standard compatibility means skills can be loaded programmatically into any agent framework, and the Apache 2.0 licence permits both commercial and non-commercial use.
Contributing
The project is actively maintained and community contributions are encouraged. New skills should follow the standard structure — YAML frontmatter with framework mappings, four-section Markdown body, supporting reference files and scripts where appropriate. Pull requests are reviewed for technical accuracy and agentskills.io compliance within 48 hours. The thinnest domains (Deception Technology and Compliance & Governance) are the highest-priority areas for new contributions.
If you use the skills library in academic research, a citation format is provided in the repository’s CITATION.cff file.
Final Thoughts
Anthropic Cybersecurity Skills is one of the most thoughtfully engineered open-source security projects to appear this year. The combination of structured skill format, progressive disclosure architecture, five-framework coverage, and broad platform compatibility addresses a real problem in a way that’s immediately practical rather than aspirational.
The 9,600 stars and 1,200 forks reflect genuine community uptake rather than just curiosity. It has been featured in the awesome-agent-skills and awesome-ai-security lists, indexed in the SkillsLLM marketplace, and integrated with the Casky.ai playground for interactive use without local installation.
If you’re using AI agents for any aspect of security work — whether that’s threat hunting, incident response, penetration testing, or detection engineering — this library is well worth adding to your setup. You’ll find everything at github.com/mukul975/Anthropic-Cybersecurity-Skills, with full documentation, contributing guidelines, and a structured index of all 754 skills.
Leave a Reply
You must be logged in to post a comment.