The open security arsenal for AI agents. 784 cybersecurity skills, scanner integrations, and a security MCP for Claude Code, Cursor, opencode, Gemini CLI, Cline, and any agentskills.io agent.

What's inside · Frameworks · Compatibility · Install · Integrations · FAQ · Attributions

🌐 English · Español · 中文 · Français · Deutsch · Português · 日本語 · 한국어 · Русский · Italiano

Why this exists

AI agents can write code and search the web, but they don't ship with the hands-on playbooks a security analyst actually relies on. Which Volatility3 plugin to run against a memory dump. Which Sigma rule catches Kerberoasting. How to trace a piece of user input all the way to a dangerous sink. The handful of good skill libraries that do cover this are scattered across different repos, formats, and frameworks.

Sectinel pulls them into one place. It bundles the strongest open security-skill libraries, wires in the leading open scanners and a security MCP, maps everything to five industry frameworks, and ships the whole thing behind one installer. Any agentskills.io-compatible agent gets all of it with nothing to assemble.

What's inside

784 skills drawn from three upstream libraries (vendored under their original licenses, with full attribution), plus the scanner and MCP wiring:

Library	Skills	Coverage
Anthropic-Cybersecurity-Skills (@mukul975)	754	26 domains: cloud, threat hunting, threat intel, web/app, network, malware, DFIR, SecOps, IAM, SOC, containers, OT/ICS, API, vuln mgmt, IR, red team, pentest, endpoint, DevSecOps, phishing, crypto, zero-trust, mobile, ransomware, compliance, deception
cybersecurity-skills (@briiirussell)	29	7 families: AppSec & supply chain, offensive/recon, detect & respond, cloud & infra, AI security, design & governance, compliance & privacy
claude-cybersecurity (@AgriciDaniel)	1 (8 agents)	flagship orchestrator that fans out to 8 parallel specialists: vuln detection, authz, secrets, supply chain, IaC, threat-intel, AI-code patterns, business logic

Browse the full catalog in docs/arsenal.md or under arsenal/.

Skill anatomy (agentskills.io standard)

Every skill is a folder with a SKILL.md: YAML frontmatter for fast discovery plus a structured Markdown workflow, often alongside references/, scripts/, and assets/. An agent reads the frontmatter (roughly 30 tokens each) and loads only the one playbook it needs (500 to 2,000 tokens), so you get full coverage without filling up the context window.

Five frameworks, one arsenal

Skills carry mappings to the main industry frameworks, so a single playbook can answer to several compliance lenses at once:

Framework	Scope
MITRE ATT&CK	adversary behaviors / TTPs (14 tactics)
NIST CSF 2.0	organizational posture (6 functions)
MITRE D3FEND	defensive countermeasures
MITRE ATLAS	AI/ML adversarial threats
NIST AI RMF	AI risk management

Compatibility

Sectinel is built on the open agentskills.io standard, so it drops into any compatible agent runtime without conversion:

Runtime	Status
Hermes Agent (NousResearch)	✅ agentskills.io skills load natively
OpenClaw	✅ reads the agentskills.io arsenal directly
opencode	✅ first-class (installer wires `~/.config/opencode`)
Claude Code	✅ first-class (`~/.claude/skills`)
Cursor · Cline · Windsurf · Roo Code · Continue · Aider	✅ via the standard and `adapters/`
OpenAI Codex CLI · Gemini CLI	✅ via the standard and `adapters/`

See adapters/README.md for per-platform drop-in files (AGENTS.md, GEMINI.md, CONVENTIONS.md, .clinerules, Cursor .mdc, and more).

Install

# Clone, then install the arsenal (skills + flagship + scanner/MCP wiring)
git clone https://github.com/Mikaru0Mystic/sectinel.git
cd sectinel
bash scripts/install.sh        # macOS / Linux / WSL
#  or, on Windows:  pwsh scripts/install.ps1

This installs the 784-skill arsenal to ~/.config/opencode/cybersec-arsenal/, the flagship cybersecurity skill to ~/.claude/skills/, and prints the security-MCP wiring snippet. Restart your agent runtime afterward.

Integrated scanners & engines

Sectinel's skills know how to drive real tools. These run at runtime rather than being bundled, and each keeps its own license (see Attributions):

Tool	Role	Bundled?
ship-safe	23-agent defensive scanner (no API key)	invoked (`npx`)
Shannon	autonomous white-box AI pentester	invoked (AGPL, not bundled)
PentAGI	autonomous pentest platform (nmap/metasploit/sqlmap)	invoked (not bundled)
Semgrep / OSV-Scanner / Trivy / Gitleaks / Checkov / hadolint	SAST / SCA / IaC / secrets	invoked
Semgrep MCP	SAST over the Model Context Protocol	`mcp/` config

See docs/integrations.md and mcp/README.md.

Pair with Breachproof (optional)

Sectinel is the arsenal, and it stands on its own with any agentskills.io-compatible agent. If you want a hands-off operator to wield it, Breachproof is an autonomous agent built around this library: it picks up the skills, runs the scanners, and drives a codebase down to zero findings. Completely optional; Sectinel needs nothing else to be useful.

Updating the arsenal

The libraries are vendored so the install stays self-contained. You can refresh them from upstream whenever you like:

bash scripts/sync-arsenal.sh

FAQ

What is Sectinel? Sectinel is an open-source security arsenal for AI coding agents. It bundles 784 cybersecurity skills, wires in open scanners and a security MCP, and maps everything to OWASP, MITRE ATT&CK, NIST CSF, D3FEND, and ATLAS, behind one installer.

Which AI agents and tools does it work with? Any agentskills.io-compatible agent: Claude Code, opencode, Cursor, Gemini CLI, Cline, Windsurf, Continue, Aider, OpenAI Codex CLI, Hermes Agent, and OpenClaw. See Compatibility and adapters/ for drop-in files.

How do I add security skills to Claude Code, Cursor, or opencode? Clone the repo and run scripts/install.sh (or scripts/install.ps1 on Windows), then restart your agent. See Install.

Is Sectinel free and open source? Yes. Sectinel's own code is Apache-2.0; the vendored libraries keep their original licenses. See License.

How is it different from a security MCP server? An MCP server exposes one tool over a protocol. Sectinel is a whole library of step-by-step security playbooks plus the wiring to drive many scanners (including a Semgrep MCP). It gives your agent the know-how, not just one endpoint.

What can it check for? OWASP Top 10, secrets, dependency and supply-chain CVEs, IaC and container misconfiguration, cloud and IAM, API and web-app flaws, malware and DFIR, incident response, threat hunting, and AI/LLM security, across 26 domains.

Does it run attacks? Not by default. Defensive and static analysis are always safe. Active-exploitation engines are opt-in and only for targets you own or are authorized in writing to test.

Contributing

The most useful contributions are field feedback and new skills. A new skill should follow the agentskills.io SKILL.md format and include framework mappings (CI checks for name and description). See CONTRIBUTING.md and the Code of Conduct.

If you're fixing a vendored upstream skill, please send the fix to the original project too (linked below) so everyone benefits, not just this copy.

License

Apache License 2.0 covers Sectinel's own code and packaging. The vendored libraries keep their original licenses (Apache-2.0 and MIT). See each library's LICENSE file under arsenal/, along with NOTICE and ATTRIBUTIONS.md.

Credits & attributions

Sectinel is really a curation and integration project. It wouldn't exist without the people who wrote the libraries and tools it brings together. Each one is their work, used under its own license:

Vendored skill libraries (bundled in `arsenal/`, with their LICENSE files)

Anthropic-Cybersecurity-Skills, 754 skills, by Mahipal Jangra (@mukul975). Apache-2.0.
cybersecurity-skills, 29 skills, by Bri Russell (@briiirussell). MIT.
claude-cybersecurity, flagship 8-agent skill, by @AgriciDaniel. MIT.

Integrated tools (invoked, not bundled)

ship-safe by @asamassekou10. MIT.
Shannon by Keygraph. AGPL-3.0.
PentAGI by vxcontrol. Apache-2.0 / EULA.
Semgrep, OSV-Scanner, Trivy, Gitleaks, Trufflehog, Checkov, hadolint.

Standards & frameworks

agentskills.io · MITRE ATT&CK · MITRE D3FEND · MITRE ATLAS · NIST CSF 2.0 · NIST AI RMF.

"Anthropic" and "Claude" are trademarks of Anthropic PBC. All other names and trademarks belong to their respective owners. Sectinel is an independent project, not affiliated with, sponsored by, or endorsed by any of the above. Provided for authorized, defensive security use only.

Sectinel

README