NanoClaw

What it is

The smallest faithful Claude-Code-shaped clone. Read this when you want to feel the irreducible structure of an agentic CLI without the production overhead.

What’s worth studying

• The generator-yield agent loop in ~200 lines. Same shape as Claude Code’s, easier to read end-to-end. Trace one iteration: prompt assembly, stream, parse, dispatch, observe, loop.
• cache_control placement at minimum scale. You can see exactly which content blocks are cacheable and where the boundary sits. Compare to Claude Code’s sentinel-string approach — same idea, smaller surface.
• Anthropic SDK only. Zero abstraction tax. Useful as a counter-example to the LiteLLM / per-adapter complexity you might add later. See provider-abstraction.

Pair this with Mistral Vibe to compare a TS+Anthropic minimal vs a Python+OpenAI-fn minimal — same idea, two language ecosystems.

Drill-down

The full per-doc analysis lives below — these are the original numbered analyses, rendered as styled HTML. Pick a section to study deeper.

• NanoClaw Architecture Overview
• Entry Points & Execution Flow
• Core Logic & Data Flow
• Key Abstractions & Design Patterns
• Configuration
• Lessons for Agentic Framework Builders
• Landscape Analysis: NanoClaw, OpenClaw, and the Agent Assistant Ecosystem
• NanoClaw Architecture Documentation
• Apple Container Networking Setup (macOS 26)
• Branch & Fork Maintenance Guidelines
• NanoClaw Debug Checklist
• NanoClaw Documentation
• NanoClaw Requirements
• Claude Agent SDK Deep Dive
• NanoClaw Security Model
• NanoClaw Specification
• Running NanoClaw in Docker Sandboxes (Manual Setup)
• Skills as Branches