AI agents are fast, but not careful. They race to an answer without checking their own logic. The result?
| Failure Mode | What Happens |
|---|---|
| Hallucination | Confidently states incorrect facts as truth |
| Surface-level answers | Fixes the symptom, not the root cause |
| Missing edge cases | Works in the demo, breaks in production |
| No self-doubt | Never says "I'm not sure" even when it should |
| Single-path thinking | Presents one option without exploring alternatives |
Deep Thinking Skill solves this by embedding a 7-phase structured reasoning protocol directly into the agent's behavior.
This is not a prompt. This is an operating system for thinking.
| Feature | Description |
|---|---|
| 🔍 Problem Intelligence | Forces the agent to restate the problem, map constraints, and surface hidden assumptions before solving |
| 🔧 Atomic Decomposition | Breaks complex problems into independently solvable, testable sub-problems |
| 🧮 Adversarial Verification | Built-in "steel-man the opposite view" step to catch logical fallacies |
| 🔀 Multi-Path Analysis | Always generates 2+ alternatives with a trade-off matrix |
| 🛡️ Failure Mode Detection | Pre-mortem analysis, blast radius assessment, and mitigation plans |
| 📊 Confidence Scoring | Self-assessed 0-100 score with hard rules (below 50% = refuse to answer confidently) |
| 🧩 12 Reasoning Frameworks | Built-in library: 5 Whys, FMEA, Porter's Five Forces, First Principles, RICE, and more |
| 🔄 Meta-Cognition | The agent checks whether it actually followed the protocol at the end |
# Install as a project skill
git clone https://github.com/shade-solutions/deep-thinking-skill.git .claude/skills/deep-thinking
# Or install as a personal skill (available across all projects)
git clone https://github.com/shade-solutions/deep-thinking-skill.git ~/.claude/skills/deep-thinkingnpx skills install shade-solutions/deep-thinking-skill# Clone and link locally
git clone https://github.com/shade-solutions/deep-thinking-skill.git
cd deep-thinking-skill
npx skills link .Once installed, the skill activates automatically when you use trigger words:
analyze • solve • optimize • strategy • why • debug • design
trade-off • evaluate • investigate • plan • root cause
"Analyze the scalability bottleneck in our microservices architecture."
"Why does our payment worker throw duplicate key errors 1 in 1000 times?"
"Design a real-time chat system that handles 10M concurrent users."
"Should we pivot from CRM to Lead Gen? We have 3 months of runway."
"Optimize our dashboard API — it takes 12 seconds to respond."
Instead of a quick (potentially wrong) answer, the agent will:
- ✅ Restate the problem to confirm understanding
- ✅ Decompose it into atomic sub-problems
- ✅ Reason through each step with hypothesis → simulate → verify
- ✅ Compare multiple solution paths with a trade-off matrix
- ✅ Detect failure modes before they happen
- ✅ Score its own confidence (and refuse to answer if too low)
- ✅ Present a structured, actionable final output
Phase 0 ⚡ ACTIVATION CHECK ─── Is this problem worth deep thinking?
Phase 1 🔍 PROBLEM INTELLIGENCE ─── Restate, define objective, map constraints
Phase 2 🔧 DECOMPOSITION ─── Break into atomic sub-problems
Phase 3 🧮 SYSTEMATIC REASONING ─── Hypothesize → Simulate → Verify → Refine
Phase 4 🔀 MULTI-PATH ANALYSIS ─── Generate alternatives, trade-off matrix
Phase 5 🛡️ FAILURE MODE DETECTION ─── Pre-mortem, blast radius, mitigations
Phase 6 📊 CONFIDENCE SCORING ─── Self-assess 0-100, hard thresholds
Phase 7 📋 STRUCTURED OUTPUT ─── Executive summary, steps, caveats
| Problem Type | Framework | When to Use |
|---|---|---|
| Root Cause | 5 Whys + Fishbone | Debugging, incident response |
| Architecture | C4 Model + ATAM | System design, tech debt |
| Decisions | Eisenhower Matrix + RICE | Prioritization, resource allocation |
| Optimization | Theory of Constraints | Bottleneck identification |
| Strategy | Porter's Five Forces + SWOT | Business/competitive analysis |
| Risk | FMEA | Safety-critical systems |
| Innovation | First Principles | Novel problem spaces |
| Trade-offs | Pareto Analysis (80/20) | Cost-benefit, feature scoping |
Real-world demonstrations of the Deep Thinking protocol in action:
| Example | Domain | Difficulty | Key Technique |
|---|---|---|---|
| 🔧 Debugging Race Conditions | Distributed Systems | Hard | TOCTOU analysis, timing simulation |
| 📊 Strategic Pivot Decision | Startup Strategy | Hard | Porter's Five Forces, RICE scoring |
| 🎲 Monty Hall Probability | Mathematics | Moderate | Bayesian proof, Monte Carlo simulation |
| 🏗️ Real-Time Chat at Scale | System Architecture | Hard | C4 Model, FMEA, back-of-envelope math |
| ⚡ 12-Second API Fix | Performance Engineering | Hard | Theory of Constraints, Pareto analysis |
deep-thinking-skill/
├── SKILL.md # 🧠 The brain — 7-phase reasoning protocol
├── metadata.json # ⚙️ Machine-readable skill metadata
├── README.md # 📖 You are here
├── CONTRIBUTING.md # 🤝 Community contribution guide
├── CODE_OF_CONDUCT.md # 📜 Community standards
├── LICENSE # ⚖️ MIT License
└── examples/
├── debugging.md # 🔧 Race condition root cause analysis
├── decision-making.md # 📊 Startup pivot strategy
├── problem-solving.md # 🎲 Monty Hall Bayesian proof
├── system-design.md # 🏗️ 10M-user chat architecture
└── optimization.md # ⚡ 12-second API performance fix
This is a community-first project. We welcome:
- 🧩 New reasoning frameworks (Socratic Method, Inversion, etc.)
- 📝 New real-world examples
- 🔧 Improvements to the SKILL.md protocol
- 🐛 Bug reports for logical gaps in examples
See CONTRIBUTING.md for details.
We believe the next generation of AI agents won't just be faster — they'll be disciplined thinkers.
Today's agents are like brilliant interns: fast, eager, but they skip steps and don't check their own work. Deep Thinking Skill is the senior engineer mindset — embedded directly into the agent's reasoning loop.
The goal: Every AI agent in the world should think before it speaks.
- 7-phase reasoning protocol (v2.0)
- Confidence scoring system
- Failure mode detection
- 12 built-in reasoning frameworks
- 5 real-world examples
- Integration with memory/retrieval skills
- Chain-of-thought visualization tool
- Community-contributed framework packs
- Automated reasoning quality benchmarks
If this project makes your AI agents smarter, consider giving it a ⭐
Built with 🧠 by shade-solutions
Think deeper. Build better.