02-core-use-jdocmunch-for-docs-retrievallisted

# Use jdocmunch for Docs Retrieval (CORE) **Why:** Whole docs/READMEs/specs are huge; you usually need one section. jdocmunch parses docs by heading into sections and returns only what matched. **BARE MODE (the user 2026-07-17):** it runs as-is on built-in defaults — NO embedding model, NO external summarizer, NO API key (the sentence-transformers + DeepSeek wiring was removed for speed; model load + API round-trips were the drag). Search is lexical (BM25/keyword) — phrase queries with the doc's own vocabulary work best; don't rely on semantic paraphrase matching. 1. **Docs → jdoc, not Read**: For any documentation/prose lookup, call a jdocmunch tool first — it returns just the section you need, where reading a whole doc to find one passage wastes context. (Native Read is also gate-removed, rule 20b.) 2. **LAZY FIRST INDEX**: Before the first lookup, `doc_list_repos`; if absent, index it: `index_local --path <folder>` (local) or `doc_index_repo --url <owner/repo>` (GitHub). Both are incremental by default. 3. **THERE'S A BUILT-IN GUIDE**: Call `jdocmunch_guide` once if unsure which tool fits — it's the server's own usage map. ## Tools & when to use **Index/manage:** `doc_index_repo` (GitHub docs), `index_local` (local folder), `index-file`/reindex one file after an edit, `doc_list_repos`, `get_index_overview` (repo at a glance), `delete_index`. **Find content (your defaults):** `search_sections` (keyword/BM25 search — main entry; no semantic layer in bare mode), `search_ti