ultraindexlisted
Install: claude install-skill maxgfr/ultraindex
# ultraindex — the verified knowledge layer over codeindex
The mechanical work is not ours. `ultraindex` vendors
**[codeindex](https://github.com/maxgfr/codeindex)** — a deterministic,
zero-dependency, keyless engine — and *it* does the scanning, symbol
extraction, import resolution and link-graph (`node scripts/ultraindex.mjs
<command>` — no `npm install`, no API keys, run `--help` for the full surface).
Everything ultraindex adds exists only because **you** are in the loop: the
encyclopedia is your durable memory of this repo — it outlives the session and
the context window — and the gates exist so nothing you write into it can be
unfounded. **If the user only needs to find code fast, say so and point them at
codeindex; they do not need this skill for that.**
> **The core rules:**
> 1. The **codeindex engine** owns the *code view* and the *graph* (`ui:gen`
> regions) — regenerated every build; never hand-edit them.
> 2. You own the *business view* (`ui:human` regions). `build` preserves your
> prose across rebuilds and renames.
> 3. **Analyze from evidence, not memory.** Write analysis only from the real
> source `dossier` shows you, cite it `[file:line]`, and `check` fails on any
> citation that doesn't resolve — so don't guess.
> 4. **Load the minimum.** Read one entry (or one `dossier`) at a time — that is
> the intended pattern, and exactly what a per-module enrichment subagent does.
> Never bulk-load `graph.json` or the whole `encyclopedia/` direct