5.8K ·GitHub
What is this?
A collection of deep-dive research reports (PDFs) analyzing the internal source code of AI Agents such as ClaudeCode and Hermes Agent, plus a small teaching-oriented Python agent project added to the same repo.
A collection of deep-dive research reports (PDFs) analyzing the internal source code of AI Agents such as ClaudeCode and Hermes Agent, plus a small teaching-oriented Python agent project added to the same repo.Is this just a report to read, or is there actual code to run?Both: most of the repo distributes PDF analysis reports, but it now also ships a minimal, runnable Python agent (src/agt/) built with Poetry purely for teaching how an AI Agent's core structure is organized.
Both: most of the repo distributes PDF analysis reports, but it now also ships a minimal, runnable Python agent (src/agt/) built with Poetry purely for teaching how an AI Agent's core structure is organized.How do I run the teaching agent?Install the project's dependencies with Poetry, then launch the minimal Agent via its CLI entry point at src/agt/cli.py.
Install the project's dependencies with Poetry, then launch the minimal Agent via its CLI entry point at src/agt/cli.py.Do I need a real LLM API key to try the teaching agent?No. The current version is a teaching-only minimal implementation that has not been wired up to a real remote model API; it ships with a swappable Fake LLM instead.
No. The current version is a teaching-only minimal implementation that has not been wired up to a real remote model API; it ships with a swappable Fake LLM instead.Does the repo include the original source code that's being analyzed (e.g. ClaudeCode's own code)?No, the repository only keeps analysis and commentary material intended for learning and review; it does not include the analyzed projects' own source directories.
No, the repository only keeps analysis and commentary material intended for learning and review; it does not include the analyzed projects' own source directories.Which AI agents does the research actually cover?The reports cover ClaudeCode in depth (v2 and v2.1, the latter adding a chapter on its memory system) and a newer deep-dive into Hermes Agent.
The reports cover ClaudeCode in depth (v2 and v2.1, the latter adding a chapter on its memory system) and a newer deep-dive into Hermes Agent.What does the minimal teaching agent focus on right now?It centers on the Agent's main loop, a Fake LLM interface, Skills discovery, and a basic CLI skeleton, kept deliberately small for readability.
It centers on the Agent's main loop, a Fake LLM interface, Skills discovery, and a basic CLI skeleton, kept deliberately small for readability.Why use a Fake LLM instead of a real model in the teaching project?So that wiring up a real model later only requires swapping the LLM call layer, without having to rewrite the rest of the Agent.
So that wiring up a real model later only requires swapping the LLM call layer, without having to rewrite the rest of the Agent.Is all of the research content free to download?Most PDF reports are linked directly for download in the repo, but at least one of the newest reports (on Hermes Agent) is only shared inside the author's paid Zhishi Planet (知识星球) community '矩阵之外'.
Most PDF reports are linked directly for download in the repo, but at least one of the newest reports (on Hermes Agent) is only shared inside the author's paid Zhishi Planet (知识星球) community '矩阵之外'.What are similar projects?harness-engineering, Gnos, oh-my-subagents, learn-coding-agent, vibe-wise, ai-guide, awesome-claude-code, vibe-coding-cn
harness-engineering, Gnos, oh-my-subagents, learn-coding-agent, vibe-wise, ai-guide, awesome-claude-code, vibe-coding-cnWhich topics is it in?cli-tool, pdf, beginner-friendly, anthropic, agents, documentation, coding-agent, claude, learning, tutorial, research, agent
cli-tool, pdf, beginner-friendly, anthropic, agents, documentation, coding-agent, claude, learning, tutorial, research, agent