A shared, living knowledge base on LLM agents — their architecture, how to evaluate them, and the papers behind them. Maintained by the Gamekeepers lab; contributed to by team members and AI agents alike.
Published with MkDocs Material → https://gamekeepers.github.io/agents-handbook/
Reading and learning evaporate unless they're consolidated somewhere durable. Every reading/learning task in the lab ends with a contribution here, so knowledge compounds across projects (Moonknight, Agentic Evaluation, …) and onboards the next person faster.
python -m venv .venv && source .venv/bin/activate
make install # pip install -r requirements.txt
make serve # live preview at http://127.0.0.1:8000
make build # strict build (fails on broken links / warnings)| Folder | Holds |
|---|---|
docs/foundations/ |
The mental model — what an agent is, the agent loop |
docs/architecture/ |
Building blocks — memory, planning, tool use, multi-agent |
docs/evaluation/ |
How we evaluate agents — methodology, metrics, LLM-as-judge |
docs/patterns/ |
Reusable design patterns / recipes we adopt |
docs/papers/ |
One structured note per paper (reading-list output) |
docs/syntheses/ |
Cross-paper writeups that tie ideas together |
docs/projects/ |
How our own projects apply the theory |
docs/glossary.md |
Single source of truth for terminology |
Humans: see CONTRIBUTING.md. AI agents: see AGENTS.md. In short — one page per idea, use the templates, link terms to the glossary, open a PR.