basemind-opencode 0.2.3 → 0.2.5

package/package.json

@@ -1,36 +1,36 @@
 {
   "name": "basemind-opencode",
-  "version": "0.2.3",
+  "version": "0.2.5",
   "description": "Full AI context layer over MCP — tree-sitter code-map, document RAG (PDF/Office/HTML/email + OCR + reranker), shared agent memory, on-demand web crawl, git history + blame + per-symbol diff. 300+ languages, 8 coding-agent harnesses, content-addressed Fjall + LanceDB.",
-  "type": "module",
-  "main": "basemind.js",
   "keywords": [
-    "mcp",
     "agent-context",
-    "rag",
     "code-map",
+    "mcp",
+    "rag",
     "tree-sitter"
   ],
+  "homepage": "https://github.com/Goldziher/basemind#opencode",
+  "bugs": {
+    "url": "https://github.com/Goldziher/basemind/issues"
+  },
+  "license": "MIT",
   "author": {
     "name": "Na'aman Hirschfeld",
     "email": "nhirschfeld@gmail.com"
   },
-  "license": "MIT",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/Goldziher/basemind.git",
     "directory": "opencode-plugin"
   },
-  "homepage": "https://github.com/Goldziher/basemind#opencode",
-  "bugs": {
-    "url": "https://github.com/Goldziher/basemind/issues"
-  },
   "files": [
     "basemind.js",
     "skills/",
     "commands/",
     "README.md"
   ],
+  "type": "module",
+  "main": "basemind.js",
   "engines": {
     "node": ">=18"
   }

package/skills/basemind/SKILL.md

@@ -8,11 +8,24 @@ description: >-
   or open many files to find structural information.
 ---
 
-# basemind — code-map MCP server
+# basemind — the indexed context layer
 
-basemind is a tree-sitter-backed code map plus git context, served over MCP. It
-pre-indexes a repository into a Fjall inverted index so structural and historical
-questions resolve in milliseconds — without you reading whole files.
+basemind is the full context layer for this repository, served over MCP. It pre-indexes the
+repo into a content-addressed blob store + Fjall inverted index (and, when enabled, a LanceDB
+vector store) so structural, historical, and semantic questions resolve in milliseconds —
+without you reading whole files.
+
+## Capabilities
+
+- **Code map across 300+ languages** — tree-sitter outlines, symbol search, references,
+  callers, call graphs, implementations, dependents.
+- **Full-text + symbol search** — indexed regex over content (`workspace_grep`) and substring
+  symbol lookup (`search_symbols`).
+- **Git intelligence** — history, blame, and structural diffs at symbol resolution, plus churn.
+- **Document RAG over 90+ file formats** — PDFs, Office, HTML, email, images (OCR) → semantic
+  search with cross-encoder reranking (`search_documents`).
+- **Shared memory** — per-repo, scope-keyed key-value + semantic memory across sessions.
+- **Web crawl** — scrape / follow-link crawl into the same searchable document store.
 
 ## When to reach for it (instead of `grep` / `read_file`)
 
@@ -29,6 +42,28 @@ If you are about to open more than two or three files just to learn structure, s
 and use basemind first. The tools return paths + line numbers; you only `read_file`
 once you know exactly which span you need.
 
+## Context economy — the operating discipline
+
+basemind tools return **paths, line numbers, and signatures — not file bodies**, so a
+structural answer costs a fraction of the tokens of reading source. Treat that as the
+default workflow, not an optimization:
+
+- **`outline` a file before you open it.** Read the whole file only when you have already
+  identified the exact span you need from the outline; then `read_file` that range, not the file.
+- **`search_symbols` instead of `grep`/`rg` for a definition.** It matches on indexed symbol
+  names and returns `path:line`, skipping the comment/string/test-name noise grep drowns you in.
+- **`find_references` / `find_callers` instead of grepping call sites.** Indexed call edges,
+  not text matches.
+- **`workspace_grep` instead of shelling out to ripgrep** when you genuinely need regex over
+  content — it runs over the in-RAM index and returns capped, structured hits.
+- **`rescan` after you edit code**, not a server reconnect. Pass `paths: [...]` to limit it to
+  the files you touched.
+- **Do not re-read a file basemind already mapped.** If the outline answered the question, stop.
+
+Rule of thumb: if a question is about _where_, _what calls_, _what shape_, _who changed_, or
+_what's indexed_, a basemind tool answers it cheaper than reading files. Reach for `read_file`
+only to see the actual implementation of a span you have already located.
+
 ## Tool routing (copy this into your mental model)
 
 | Question | Tool |