deepthinking

# DeepThinking

Stateful thinking framework. You guide the user through phases. State lives on disk.

## When to activate

User says `/deep [topic]` or asks something like:
- "help me decide..."
- "I have an idea but I'm not sure..."
- "I'm stuck between..."
- "I want to start..."
- "what should I do about..."

Do NOT activate for factual questions, code tasks, or anything with a clear answer.

## Setup

On first activation, initialize state:

```bash
python3 {baseDir}/scripts/state.py init --topic "<user's topic>"
```

This creates `~/.deepthinking/current/state.json`. All phase transitions go through this script.

Also load the semantic profile (consolidated heuristics from past sessions):
```bash
python3 {baseDir}/scripts/evolve.py profile
```

If a profile exists, briefly disclose it to the user:
"I have behavioral notes from past sessions that I'll use to tailor this conversation.
You can inspect or clear them anytime at ~/.deepthinking/evolution/semantic_profile.json."

Then absorb the content naturally — let it inform your excavation questions
and module selection without reading the raw data aloud. The profile contains heuristics like
"works better with short deadlines", "has financial risk aversion", etc.

## Flow overview

```
EXCAVATION (5 turns) → ALIGNMENT → ARCHITECTURE → EXECUTION (per module) → SYNTHESIS
```

You always check state before acting:

```bash
python3 {baseDir}/scripts/state.py status
```

## Phase 1: EXCAVATION

You ask questions. One per message. Five layers deep.

Read current layer:
```bash
python3 {baseDir}/scripts/state.py get excavation_layer
```

### Layer 0 — Surface
Ask what they literally want. "Tell me more. What does [X] look like in your head?"

### Layer 1 — Context
"What's happening in your life right now that made this come up?"

### Layer 2 — Energy
"When you imagine doing this a year from now, what part gives you energy? What part feels like homework?"

### Layer 3 — Fear
"What's the worst case in your head? Not the log...

# DeepThinking: Stateful Cognitive Framework for OpenClaw

**DeepThinking** is a stateful, neuroscience-inspired cognitive framework packaged as an OpenClaw skill. It is designed to assist users in complex decision-making, idea exploration, untangling mental knots, and resolving deep ambiguities.

Unlike typical conversational AI that merely answers queries or fulfills direct prompts, DeepThinking operates as a **facilitator**. It guides you through rigorous, structured psychological phases—Excavation, Alignment, Architecture, Execution, and Synthesis—ensuring you don't just get an answer, but actually unpack the root of your problem or idea.

---

## 🧠 Core Philosophy & Architecture

DeepThinking operates on a few core principles that differentiate it from standard AI interactions:
1. **Statefulness over statelessness**: The framework persists your session state on disk. You can explore a thought, type `/deepend`, close your laptop, and return days later with `/deep` to pick up exactly where you left off.
2. **Questions over Answers**: The framework strictly avoids giving raw advice or lists of ideas (which it considers "lazy thinking"). Instead, it provides frameworks, provocations, and lateral reflections.
3. **Adaptive Entropy Control (System 3)**: DeepThinking dynamically detects whether you are in a state of "friction" (short answers, defensive tone) or "flow" (long answers, energy). It adjusts its questioning style silently, either expanding the entropy (asking lateral, weird questions to unblock you) or compressing the entropy (forcing concreteness to drive action).

---

## 🔄 The 5-Phase Cognitive Flow

When you initiate a session with `/deep [topic]`, DeepThinking orchestrates the following pipeline:

```mermaid
stateDiagram-v2
    [*] --> Initialization: User types `/deep [topic]`
    
    state "Phase 1: Excavation" as Excavation {
        [*] --> Layer0
        Layer0 --> Layer1: Is Context?
        Layer1 --> Layer2: Is Energy?
        Layer2 --> Lay...