Claude Ecosystem Wiki

claude-ecosystem-wiki

An **agent-maintained wiki**. Every page under `wiki/` was written and is kept current by [Claude Code](https://code.claude.com), following the schema in [`CLAUDE.md`](CLAUDE.md). A human curates the sources and asks the questions; the agent does the summarising, cross-referencing, filing and bookkeeping.

The subject happens to be the Claude ecosystem itself — models, safeguards, MCP, skills, plugins, and the policy that governs access to them. But the *method* is the transferable part, and re-pointing it at another domain means rewriting one section of `CLAUDE.md`.

---

The idea

Most LLM document workflows are retrieval: upload files, ask a question, get relevant chunks assembled on the spot. Nothing accumulates — the same connective work is redone on every query.

This is the alternative. When a source arrives, the agent reads it and **integrates** it: writes a summary page, updates every entity and concept page it touches, flags where it contradicts what's already there, and revises the standing thesis. The cross-references are written down. The contradictions are already surfaced. Ask a question later and the synthesis is waiting rather than being reassembled.

The bet is that maintenance — not reading, not storage, not retrieval — was always the binding constraint on personal knowledge bases, and that it is the one an agent removes.

Layout

CLAUDE.md      the schema: rules, conventions, and the three operations
index.md       catalog — every page, one line each
log.md         append-only record of every ingest, query and lint
raw/           source documents (withheld — see below)
wiki/          the knowledge pages; the agent owns every file here
  overview.md    start here
  synthesis.md   current theses, with confidence ratings
  sources/       one page per source
  entities/      named things — models, products, companies
  concepts/      ideas, mechanisms, patterns
  questions/     answers worth keeping, filed rather th