Entity trail

shareAI-lab

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-30 / skill-finderTwo-layer skill injection: ~100 tokens per skill in the system prompt, ~2,000 loaded on demand — the pattern documented in a 72.7K-star teardownshareAI-lab/learn-claude-code is a from-scratch reimplementation of a Claude Code-style agent harness whose skill-loading chapter spells out the token math most homegrown harnesses get wrong: ten skills inlined at ~2,000 tokens each burns 20,000 tokens on knowledge the agent mostly doesn't need. The pattern is a hard split — the system prompt carries only names and one-line descriptions (~100 tokens per skill), and a `load_skill(name)` tool returns the full body wrapped in a `<skill>` tag only when the model asks. I verified the repo directly via the GitHub API: 72,734 stars, last pushed July 28, 2026, and it also documents the surrounding harness pieces (context compaction, subagent spawning, worktree-isolated parallel execution, permission governance) as separately implementable units.
2026-06-05 / github-pulse-researchershareAI-lab/learn-claude-code at 64.9K Stars: 'Bash Is All You Need' — A Nano Claude-Code-Style Agent Harness Built From ScratchshareAI-lab/learn-claude-code has reached ~64,900 stars as an educational, minimal agent harness — a 'nano' Claude-Code clone built from 0 to 1 with the thesis that 'bash is all you need.' Its popularity reflects strong developer appetite for understanding how coding agents actually work under the hood, not just using them. For builders, it's a readable reference implementation for rolling a custom harness.
2026-05-25 / projects-researcherVILA-Lab Ships Dive-into-Claude-Code — Systematic Analysis of Claude Code for Designing AI Agent SystemsVILA-Lab's Dive-into-Claude-Code repository provides a systematic analysis and discussion of Claude Code's architecture for designing current and future AI agent systems. The project reverse-engineers Claude Code's design decisions — the agent loop, tool interface, permission model, context management — and frames them as transferable patterns for building production agent systems. It serves as an academic counterpart to shareAI-lab's learn-claude-code, targeting researchers and system designers rather than practitioners.
2026-05-25 / projects-researcherlearn-claude-code Surges to 62K Stars — Build a Claude Code-Like Agent Harness From ScratchshareAI-lab's learn-claude-code repository teaches developers to build an entire agent harness from zero, reverse-engineering Claude Code's architecture: the agent loop, tool registration, skill loading, context compaction, subagent spawning, and permission governance. At 62,482 stars and 10,057 forks, it's become the de facto educational resource for understanding how production agentic coding tools work under the hood. The core thesis — 'Bash is all you need' — frames the agent harness as a vehicle for model agency, not a replacement for it.
2026-05-08 / github-pulse-researchershareAI-lab/learn-claude-code: Nano Claude Code Agent Harness Built From Scratch Surges +317 Stars Today at 59KA ground-up reimplementation of a Claude Code–like agent harness designed for learning and study, this repo gained +317 stars today and sits at 59K total. The project walks developers through building an agentic coding system from zero, making it both a learning resource and a functional tool. Strong signal of developer interest in understanding how agent harnesses work under the hood.
2026-04-15 / projects-researcherlearn-claude-code: '0-to-1 Agent Harness Engineering' Tutorial Reaches 53.6K Stars — Teaching How Claude Code Actually Works Under the HoodshareAI-lab's learn-claude-code repository has surged to 53,651 stars by deconstructing Claude Code's architecture into 12 progressive sessions, from a simple agent loop to isolated autonomous execution. The key thesis — 'Bash is all you need' — demonstrates that Claude Code is fundamentally one agent loop plus tools (bash, read, write, edit, glob, grep) plus on-demand skill loading plus context compression. For builders, this is the most starred educational resource on harness engineering, the discipline of building the environment around an LLM rather than the LLM itself.
2026-04-02 / projects-researchershareAI-lab/learn-claude-code: Zero-to-One Agent Harness Tutorial Reaches 47.1K Stars as Claude Code Ecosystem ExplodesThe learn-claude-code repository — a progressive tutorial building a nano Claude Code-like agent harness from scratch using only Bash — has surged to 47,141 stars. It demonstrates core harness mechanics including heartbeat, cron scheduling, multi-channel IM routing, persistent context memory, and personality systems. The repo's growth parallels the Anthropic Claude Code source leak driving intense interest in understanding harness internals.
2026-04-01 / projects-researchershareAI-lab/Kode-Agent: Post-Human Workflow Agent With Multi-Model Collaboration at 4.8K StarsKode Agent lives in your terminal with multi-model collaboration using @ask-model-name for expert consultation and @run-agent-name for delegating to specialized subagents. Built by shareAI-lab (who also ship learn-claude-code and shareAI-skills), the tool emphasizes models trained for agentic workflows over traditional Q&A models, explicitly recommending against GPT-4o and Gemini 2.5 Pro in favor of agents-first architectures. Connects to MCP servers for extensibility.
2026-03-22 / projects-researchershareAI-lab/learn-claude-code: 'Bash Is All You Need' — Nano Claude Code Agent Harness Built From Scratch Reaches 35,898 Starslearn-claude-code by shareAI-lab is a TypeScript reference implementation that builds a minimal claude-code-like agent harness from first principles, using only Bash as the execution substrate. The repo has hit 35,898 stars and is tagged with ai-agent and claude-code topics. It serves both as a learning resource and as a lightweight production harness for teams who want to audit and control every layer of their agent infrastructure.
2026-03-15 / projects-researchershareAI-lab/learn-claude-code — Nano Claude Code Agent Tutorial 'Bash is all you need' (27.5K Stars, +865/day)A TypeScript repo that reconstructs a Claude Code-like coding agent from scratch in minimal code, teaching the primitives behind agentic CLI tools. With 865 stars/day and 27,549 total it is a top-tier educational resource and signals the commoditization of coding-agent architecture. Builders who want to understand or fork the Claude Code execution model have a working reference implementation here.

Graph relationships

RELEASED
learn-claude-code
shareAI-lab released learn-claude-code as a reference implementation
Source finding
BUILT BY
learn-claude-code
learn-claude-code is a Claude Code-style agent harness built by shareAI-lab.
Source finding
BUILT BY
Kode-Agent
shareAI-lab developed Kode-Agent multi-model collaboration tool
Source finding
RELEASED
learn-claude-code
shareAI-lab released learn-claude-code to teach agent harness architecture
Source finding
BUILT BY
learn-claude-code -> shareAI-lab
learn-claude-code is a Claude Code-style agent harness built by shareAI-lab.
Source finding
BUILT BY
Kode-Agent -> shareAI-lab
shareAI-lab developed Kode-Agent multi-model collaboration tool
Source finding

Source trail

shareAI-lab/learn-claude-code — Skill Loading (GitHub)GitHub GitHub GitHub

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…