Concept-level field guides to open-source codebases — why each subsystem exists, how the pieces relate, and where to look in the code.
Programmatic prompting — Signature, Module, Adapter and the LM layer, plus caching, compilation, the proposer, and streaming.
An autonomous pentest app on the OpenAI Agents SDK — the agent graph, the Docker sandbox, and the finding-to-report pipeline.
A durable agent runtime — the superstep engine and version-trigger mechanism, how state compiles to channels, control flow, and what checkpoints store.
A stateful agent harness — the turn lifecycle and per-conversation queues, memory blocks, the git-tracked memory filesystem, dreaming, and client-side tools.
A browser-automation agent — the event bus and watchdogs, the CDP execution layer, the action registry, and how the agent sees a page.
Multi-agent orchestration — kickoff and the agent-executor loop, the hierarchical process, the flow scheduler, and where state actually lives.
An LLM app platform and its Graphon workflow engine — the graph execution model, the variable system, pause/resume, the model runtime, and plugins.
A workflow-automation engine — how it decides what runs next, why the canvas is not the execution, partial executions, items and pairedItem, triggers.
A deep-document RAG engine — the ingestion pipeline and chunking templates through the embedding layer, GraphRAG, and the canvas orchestrator.
Every guide opens with a system map and an end-to-end “anatomy of a call” trace, so you can hold the architecture in your head before opening a file.
Every claim cites real files and symbols, and every page ends with a “where to look in the code” map. Nothing here is folklore — it was read out of the code.
Each project’s own docs own tutorials and reference. These guides cover the connective architecture the reference pages leave implicit.
The Library is the demo. We pointed Doc Holiday at other people's open source repos and let it read the commits, PRs and issues. Point it at yours and it opens a pull request with the docs already written. You review it, you merge it, you go back to writing code.
Free on every public repo. You bring your own API key, and that is the entire bill.
