Cybersecurity Skills (mukul975) — source summary
An 817-skill cybersecurity skill pack for AI agents, built on the agentskills.io standard (by Mahipal Jangra / @mukul975; Apache-2.0; ~18.7k stars). Delivered via Telegram, ingested 2026-06-23. Like agentic-seo-skill, the security domain is incidental — the skill-pack form factor is why it’s here: this is the largest single skillpack instance the wiki has seen, and a stress test of the standard’s scaling claims.
What it is
Each “skill” is a self-contained security workflow in the standard layout: ~30-token YAML frontmatter
(name, description, domain, tags, framework mappings) + markdown body (When to Use / Prerequisites /
Workflow / Verification) + reference files (standards.md, workflows.md) + Python helpers and
checklists. Examples: performing-memory-forensics-with-volatility3, hunting-for-credential-dumping-lsass.
The 817 skills span 29 security domains (Cloud Security 66, Threat Hunting 58, Threat Intel 52, …,
down to Hardware/Firmware 4) and install across 26+ platforms that read agentskills.io (Claude Code,
Copilot, Cursor, Codex CLI, Gemini CLI, plus agent frameworks like LangChain/CrewAI/AutoGen). One command:
npx skills add mukul975/Anthropic-Cybersecurity-Skills.
Why it’s interesting here
- Progressive disclosure at scale. It is the concrete proof of agentskills-spec‘s core bet: an agent scans all 817 frontmatters in one pass (~30 tokens each), then loads only the relevant 500–2,000-token workflow. 817 skills is the largest test of “metadata → body → resources on demand” the wiki has paged — a scaling-validation data point for the standard, not just another skillpack.
- Domain specialization, but with an AI-security twist. Beyond classic security frameworks (MITRE ATT&CK, NIST CSF 2.0, D3FEND), two of its six framework mappings — MITRE ATLAS (adversarial ML / agentic-AI attack vectors) and NIST AI RMF — mean an agent skillpack ships skills for attacking and defending agents themselves (LLM red-teaming, prompt injection, agentic security: the 14-skill “AI Security” domain). The tooling and its threat model converge.
- Attribution caveat (worth flagging). Despite the repo name “**Anthropic-**Cybersecurity-Skills,” it is an independent community project, explicitly not affiliated with Anthropic — contrast the genuinely first-party anthropic-skills. A naming choice that borrows the vendor’s brand; recorded as a provenance note, not an endorsement.
Related
agentskills-spec · agent-skills · anthropic-skills · agentic-seo-skill · pm-skills · mahipal-jangra