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 ยท Frameworks ยท Compatibility ยท Install ยท Integrations ยท FAQ ยท Attributions
๐ English ยท Espaรฑol ยท ไธญๆ ยท Franรงais ยท Deutsch ยท Portuguรชs ยท ๆฅๆฌ่ช ยท ํ๊ตญ์ด ยท ะ ัััะบะธะน ยท 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:
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) | โ
agentskills.io skills load natively |
| OpenClaw | โ
reads the agentskills.io arsenal directly |
| opencode | โ
first-class (installer wires ~/.config/opencode) |
| Claude Code | โ
first-class (~/.claude/skills) |
| Cursor ยท Cline ยท Windsurf ยท Roo Code ยท Continue ยท Aider | โ
via the standard and adapters/ |
| OpenAI Codex CLI ยท Gemini CLI | โ
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)
Integrated tools (invoked, not bundled)
Standards & frameworks
agentskills.io ยท MITRE ATT&CK ยท MITRE D3FEND ยท MITRE ATLAS ยท NIST CSF 2.0 ยท 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.
