This directory holds material for human contributors working on Hecatoncheires: the reasoning behind internal design and how to extend it. If you are setting up the project locally, start with Getting Started.
Hecatoncheires keeps three distinct kinds of "rules and context" in three distinct places. Put new material where it belongs:
| Location | Audience | Nature |
|---|---|---|
.claude/rules/*.md |
Claude / agents | Enforced rules — breaking them breaks the system (architecture invariants, repository write contract, keyboard/IME policy, …) |
CLAUDE.md (root) |
Claude / all agents | Project-wide AI context, kept lightweight |
docs/develop/ (this dir) |
Human contributors | "Why we designed it this way" and "how to extend it" |
When you are unsure: if violating it would corrupt data or behavior, it is a
rule (.claude/rules/). If it is background an agent should always carry, it is
CLAUDE.md. If it is explanatory prose for a person, it belongs here.
- Architecture (internals) — the "why/how" of internal design: GraphQL DataLoader (request-scoped batching) and the Agent thread session implementation (Cloud Storage layout, IAM, turn-lock / state persistence).
- Concepts — product vocabulary
- Configuration and CLI Reference
- Root README and CLAUDE.md