ARC-AGI-3 Cognitive Architecture — 6 human drives, 9 reasoning modes, SkillDAG orchestration, and human-like skill acquisition from zero (drawing fundamentals → writing → reading → painting).
Discovers then solves. No brute force, no templates — an architecture that generalizes.
| Track | Repository | Focus |
|---|---|---|
| Cognitive Solver | cogniarc/ (this repo) |
ARC-AGI-3 puzzle solving via 6 drives + 9 reasoning modes + dynamic workflows |
| Human Skills | arc-human-skills/ |
Learn to read, write, paint like a human from zero — Watch tutorials → Practice in MS Paint → Self-evaluate → Transfer skills |
Both share the SkillDAG architecture (atomic skills + topological dependencies) for composable, transferable learning.
CogniARC Solver
├── Drives (6)
│ ├── Curiosity — Explore the unknown
│ ├── Pattern_Match — Recognize visual patterns
│ ├── Causal — Understand cause-effect
│ ├── Efficiency — Seek simplest solution
│ ├── Memory — Retain past patterns
│ └── Verify — Validate and self-correct
├── Reasoning Modes (9)
│ ├── Symbolic — Symbol manipulation
│ ├── Spatial — Spatial reasoning
│ ├── Temporal — Sequential reasoning
│ ├── Analogical — Analogy & transfer
│ ├── Causal — Causal chains
│ ├── Compositional — Pattern composition
│ ├── Relational — Object relations
│ ├── Probabilistic — Uncertainty estimation
│ └── Meta — Reasoning about reasoning
└── Dynamic Workflows (6)
├── Classify and Act — Type classification → strategy
├── Fan Out & Synthesize — Parallel hypotheses
├── Adversarial Verification — Skeptic agent review
├── Generate and Filter — N candidates → best
├── Tournament — Pairwise comparison
└── Loop Until Done — Iterate to verification
Learn to READ, WRITE, and PAINT like a human — from absolute zero — using Windows Paint, video tutorials, and iterative self-evaluation.
┌─────────────────────────────────────────────────────────────┐
│ SKILL DAG ORCHESTRATOR │
│ (Topological scheduling, prerequisites, mastery tracking) │
└──────────────────────┬──────────────────────────────────────┘
│
┌──────────────┼──────────────┐
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ DRAWING │ │ WRITING │ │ READING │
└────┬────┘ └────┬────┘ └────┬────┘
│ │ │
└──────────────┼──────────────┘
▼
┌─────────────────────┐
│ PAINTING │
│ (Bob Ross style) │
└─────────────────────┘
| Level | Skills | Description |
|---|---|---|
| 0 — Line Control | 13 | Horizontal, vertical, 4 diagonals, pressure fade in/out/wave |
| 1 — 2D Primitives | 9 | Cross, square, rectangle, X, diamond, △, hexagon, octagon, circle |
| 2 — 3D Wireframes | 8 | Cube/box (iso/1pt/2pt), pyramid, prism, cylinder, cone |
| 3 — Perspective | 7 | Horizon/VPs, grids (ground/wall), ellipses, shadows, measuring |
| 4 — Construction | 3 | Still life, interior corner, building exterior |
Total: 39 drawing skills + 12 writing + 4 reading + 12 painting + 7 transfer = 74 atomic skills in unified DAG.
- Embodied practice — Real drawing in MS Paint via pywinauto automation
- Self-supervision — LocalAI vision (qwen3.6-27b) evaluates own output
- Geometric evaluation — Angle tolerance ±3°, length ±5%, closure <5px (no vision needed for basics)
- Cross-domain transfer — Strokes→Letters, Primitives→Shapes, Perspective→Scenes
- Continuous learning — SkillDAG unlocks skills via mastery (5 attempts, avg ≥80%)
- Tutorial-driven — YouTube → Whisper transcription → Key frame extraction → Practice
cd C:\Users\redga\projects\arc-human-skills
run_windows.bat
# Or PowerShell with options:
.\run_windows.ps1 -Mode train -MaxSessions 10 -Duration 30 -Domains drawing,writing,readingcd ~/projects/arc-human-skills
python -m arc_human_skills.trainer --headless --max-sessions 1 --domains drawing
python -m arc_human_skills.benchmarkcd ~/projects/cogniarc
pip install -e .
python -m cogniarc.arc_agent --task <arc_task.json>| Component | Purpose |
|---|---|
| Windows 10/11 | MS Paint automation (pywinauto) |
| Python 3.11+ | Runtime |
| LocalAI on EUREKAI (192.168.1.47:8080) | qwen3.6-27b (vision), whisper-1 (STT), tts-1 (TTS) |
| Qdrant on EUREKAI (192.168.1.47:6333) | Vector embeddings for letter templates |
| NVIDIA GPU | For LocalAI inference |
cogniarc/ # Cognitive solver (this repo)
├── cogniarc/
│ ├── arc_agent.py # Main ARC solver entry point
│ ├── scientist_agent.py # Hypothesis generation (legacy)
│ ├── skill_dag/ # SkillDAG v2 (17 atomic skills)
│ ├── triarchic_engine.py # Sternberg WICS model
│ ├── representation_engine.py # Visual representation
│ ├── cognitive_player.py # Game playing interface
│ └── *_inference.py # Transform, goal, domain, causal
├── tests/
└── README.md
arc-human-skills/ # Human skills (separate repo)
├── arc_human_skills/
│ ├── config.py # Typed configuration
│ ├── paint_automation.py # Windows Paint control
│ ├── video_tutorial.py # YouTube → Whisper → keyframes
│ ├── trainer.py # Main training loop (CLI)
│ ├── benchmark.py # ARC-style evaluation
│ ├── drawing_fundamentals/ # Levels 0-4 (NEW)
│ │ ├── line_control.py # Level 0: motor control
│ │ ├── primitives_2d.py # Level 1: 2D shapes
│ │ ├── wireframe_3d.py # Level 2: 3D wireframes
│ │ ├── perspective.py # Level 3: perspective
│ │ ├── construction.py # Level 4: scenes
│ │ └── curriculum.py # Orchestrator + SkillDAG
│ ├── reading/ # Letter recognition
│ ├── writing/ # Stroke patterns + letters
│ ├── painting/ # Shapes + Bob Ross landscapes
│ ├── skill_dag/ # Unified manifest (74 skills)
│ └── eval_tasks/ # Benchmark tasks
├── tests/ # 68 passing (10 skipped on Linux)
├── run_windows.bat / .ps1 # Windows launchers
└── README.md
| Repo | Description |
|---|---|
| hermes-brain | Hermes Agent architecture |
| arc-human-skills | Human skills track (drawing/writing/reading/painting) |
| hermes-fusion | Multi-LLM fusion engine (4 providers, 4 strategies) |
MIT — See LICENSE for details.
Built for ARC-AGI-3 — Advancing human-like skill acquisition and cognitive generalization.