| name | using-sentinel |
| description | Entry for security tasks (audit, pentest, CTF, CVE). After audit/pentest/frontend audit completes, MUST load sentinel-report/SKILL.md and Write .md + .html — never stop at chat-only findings. |
Using Sentinel
Multi-host: This file is plain Markdown with YAML frontmatter. Use it from Claude Code, Cursor, Codex, Trae, or any agent that can load project rules or skill files. The prefix sentinel: in examples is a logical id — your client may expose it as a rule name, plugin skill, or file path.
Sentinel is a standalone security skill suite. It follows the same discipline as leading agent workflow packs: if a security skill might apply, invoke it. No exceptions.
Audit / pentest completion — you MUST chain sentinel-report
When sentinel-audit, sentinel-pentest, or sentinel-frontend work reaches findings delivered or audit complete:
- Open
skills/sentinel/sentinel-report/SKILL.md (path relative to this repo).
- Follow its Closing gate and
Write sentinel-security-assessment.md + sentinel-security-assessment.html.
- Do not treat “list of vulnerabilities in chat” as the final step.
If the host does not auto-load skills, you still perform the same steps (questions + file writes) — the skill file is the spec.
Instruction Priority
- User's explicit instructions — highest priority
- Sentinel security skills — override defaults for security tasks (
sentinel-audit, sentinel-frontend, sentinel-pentest, sentinel-ctf, sentinel-cve, sentinel-report)
sentinel-workflow — systematic debugging, verification, structured review, and planning (see below)
- Default system behavior — lowest priority
Skill Selection
Step 1: Detect Architecture/Technology Stack FIRST
Before invoking any skill, identify what you're working with:
☐ What is the application type?
├── Web (Frontend) → Vue.js / React / Angular
├── Web (Backend) → PHP / Python / Node.js / Java / Go / C#
├── Mobile → Android (Kotlin/Java) / iOS (Swift/ObjC)
├── API-only → REST / GraphQL / gRPC
├── Desktop → Electron / Qt / .NET
└── Infrastructure → Docker / Kubernetes / Cloud configs
☐ What is the framework (if any)?
├── PHP: Laravel / Symfony / CodeIgniter / WordPress
├── Python: Django / Flask / FastAPI
├── Java: Spring / Struts / Jakarta EE
├── Node.js: Express / Koa / NestJS / Next.js
└── Go: Gin / Echo / Fiber
☐ What is the deployment environment?
├── On-premise / Cloud (AWS/GCP/Azure)
├── Containerized (Docker/Kubernetes)
└── Serverless
Step 2: Route Based on Architecture
| Architecture Detected | Primary Skill | Reference Files |
|---|
| Frontend (Vue/React/Angular) | sentinel:sentinel-frontend | patterns-vue.md, patterns-react.md, patterns-angular.md |
| Backend (PHP) | sentinel:sentinel-audit | patterns-php.md |
| Backend (Python) | sentinel:sentinel-audit | patterns-python.md |
| Backend (Node.js) | sentinel:sentinel-audit | patterns-js.md |
| Backend (Java) | sentinel:sentinel-audit | patterns-java.md |
| API (REST/GraphQL) | sentinel:sentinel-audit | + IDOR patterns |
| Full Stack | sentinel:sentinel-audit | + frontend + backend refs |
| Live System (authorized) | sentinel:sentinel-pentest | — |
| CTF Challenge | sentinel:sentinel-ctf | — |
| Report Writing | sentinel:sentinel-report | — |
| Root cause, verification, structured review, parallel targets | sentinel:sentinel-workflow | — |
Skill Routing Flow:
digraph sentinel_routing {
"User Request" -> "Detect Architecture"
"Detect Architecture" -> "Frontend?" [label="Vue/React/Angular?"]
"Detect Architecture" -> "Backend?" [label="PHP/Python/Node/Java?"]
"Detect Architecture" -> "API?" [label="REST/GraphQL?"]
"Detect Architecture" -> "Mobile?" [label="Android/iOS?"]
"Frontend?" -> "sentinel:sentinel-frontend"
"Backend?" -> "sentinel:sentinel-audit"
"API?" -> "sentinel:sentinel-audit"
"Mobile?" -> "sentinel:sentinel-audit" [label=" (add mobile refs)"]
}
When in doubt: Invoke sentinel:sentinel-audit first — it covers the broadest ground and includes cross-site/framework patterns.
Core Security Mindset
Before invoking any sentinel skill, adopt this posture:
- Think like an attacker. What would a malicious actor target first?
- Trust nothing. Every input, every boundary, every assumption is suspect.
- Evidence over intuition. Document what you find, not what you feel.
- Scope matters. Never exceed authorized scope. Confirm before acting.
Built-in workflow (no external plugin required)
- Use
sentinel:sentinel-workflow for root-cause tracing, pre-completion verification, security-focused review of diffs, design exploration, and parallel multi-target analysis
- Use
sentinel:sentinel-report at the end of any engagement that needs a structured write-up — default write both .md + offline .html; must ask which folder to save into; user can opt out of HTML explicitly; do not stop at chat-only markdown unless the user refused files
Optional: if your environment also loads the Superpowers plugin, you may use those workflow skills interchangeably with sentinel-workflow for the same class of tasks.
Red Flags
Stop and re-read the appropriate skill if you catch yourself:
- Guessing at vulnerabilities without evidence
- Skipping scope confirmation before active testing
- Reporting a finding without a verified reproduction path
- Mixing exploit development with authorized scope