Makes AI coding agents think critically, reason deeply, and act with integrity. No hallucination. No excuses. No sycophancy.
npx skills add vtino17/deep-mind-skillFor Claude Code:
/plugin marketplace add vtino17/deep-mind-skillOnce installed, just include any of these in your prompt:
Commands:
/think-deeper <question>— Full 8-stage reasoning/critical <claim>— Adversarial review/verify <statement>— Evidence gate/deep-dive <problem>— Root cause analysis/first-principles <concept>— Break down from fundamentals
Natural language:
- "think deeper: why does this keep failing?"
- "critical analysis: is this architecture correct?"
- "verify: does this library support Python 3.13?"
User: /think-deeper why is port 443 open on 10.0.0.1?
Agent:
[DEEP-MIND] Applying the reasoning pipeline...
1. VERIFY — Port 443 open means HTTPS service running
2. GROUND — Nmap scan result: 443/tcp open https
3. REASON — HTTPS requires TLS certificate + web server
4. CHALLENGE — Could be non-HTTP service masquerading on 443
5. DECIDE — Confirmed: TLS handshake successful, subject: *.example.com
- 8-Stage Reasoning: research, clarify, deconstruct, analyze, synthesize, verify, self-correct, challenge
- Anti-Hallucination: Blocks fabricated code and imaginary APIs
- Evidence Gate: Every claim must cite real sources
- Anti-Sycophancy: Rejects user flattery, tells the truth
- Zero Dependencies: Pure markdown, no pip/npm install needed
- Agent-Agnostic: Works with Claude Code, Codex, Cursor, Windsurf, Aider, Cline, Continue, Roo Code, Gemini CLI, Copilot CLI, OpenCode
- Plugin support: OpenCode slash commands
When the skill is selected, it applies eight stages:
0. RESEARCH Verify uncertain and time-sensitive facts
1. CLARIFY Bound the request and expose assumptions
2. DECONSTRUCT Split the problem into testable parts
3. ANALYZE Compare evidence, risks, and alternatives
4. SYNTHESIZE Build a coherent answer from verified parts
5. VERIFY Test claims, outputs, and edge cases
6. SELF-CORRECT Review and repair material weaknesses
7. CHALLENGE Apply the strongest credible counterargument
If any stage fails, the agent reports the gap instead of guessing.
| Agent | Distribution included | Invocation |
|---|---|---|
| Claude Code | Yes | Skill or slash commands |
| Cursor | Yes | Skill reference |
| OpenCode | Yes | Plugin commands and skill reference |
| Codex CLI | Generic skill copy | Skill reference |
| Windsurf, Copilot CLI, Cline, Continue, Aider, Roo Code | Installer target | Manual skill reference; verify against the installed agent version |
Packaging support does not guarantee identical runtime behavior across agents.
| Repo | Stars | Description |
|---|---|---|
| kage | AI-powered security scanner | |
| tools | 85+ penetration testing tools | |
| taskcapsule | Task context manager | |
| network-security-lab | Enterprise network security lab | |
| mikrotik-hardening | RouterOS security configs | |
| incident-response-playbooks | IR playbooks and scripts | |
| vuln-scanner | Vulnerability scanner | |
| pcap-forensics | PCAP analysis tool | |
| network-automation-toolkit | Multi-vendor automation |
MIT — use, modify, share freely.