メニュー
Search, query, and explore the Lumen knowledge base via MCP tools. TRIGGER when user invokes /lumen or asks about their knowledge base, sources, concepts, or graph.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Diagnose Lumen CLI or MCP issues — common errors, SQLite problems, search returning no results, compilation failures. TRIGGER when user reports a bug, error, or unexpected behavior.
Pre-deploy verification for Lumen monorepo — type check, lint, test, build across all workspaces. TRIGGER when user says deploy, ship, release, pre-merge, or CI check.
Reference for navigating the Lumen turborepo — workspace commands, where things live, cross-app imports. TRIGGER when user asks about project structure, where to put files, or how workspaces connect.
Review changed code for Lumen project conventions — catches gotchas that linters miss like wrong hash function, FTS5 quoting, missing .js extensions, interface instead of type. TRIGGER when user asks to review, check, or audit code they wrote.
Scaffold a new page or component for the Lumen Next.js web app. TRIGGER when user asks to create, add, or build a new page, route, or dashboard section.
Scaffold a new Lumen CLI command following the Commander.js register pattern. TRIGGER when user asks to create, add, or implement a new CLI command or subcommand.
SOC 職業分類に基づく
| name | lumen-mcp |
| description | Search, query, and explore the Lumen knowledge base via MCP tools. TRIGGER when user invokes /lumen or asks about their knowledge base, sources, concepts, or graph. |
You have access to a local Lumen knowledge base via MCP tools. All data is stored in SQLite on the user's machine.
| Tool | When to use |
|---|---|
search | Find content by keyword — returns ranked chunks with snippets |
query | Ask a question — synthesizes an answer from chunks via LLM |
status | Quick stats — source count, chunk count, concept count, edges |
god_nodes | Find the most connected concepts (highest edge count) |
concept | Deep dive into one concept — summary, edges, sources |
path | Find how two concepts connect (shortest path) |
neighbors | Explore around a concept within N hops |
pagerank | Rank concepts by structural importance |
communities | List topic clusters detected by label propagation |
community | Members of a specific cluster by ID |
add | Ingest a new source (URL, file, arXiv, YouTube) |
search first, then query if user wants synthesisgod_nodes or pagerankpathneighborscommunities, then community for detailaddsearch returns raw chunks. query calls the LLM to synthesize — it costs API tokens.community IDs are ephemeral — recomputed on each call. Don't cache them across sessions.add deduplicates by content hash. Adding the same URL twice is safe.path returns null if no path exists — concepts may be in disconnected components.god_nodes and pagerank return empty arrays if lumen compile hasn't been run yet.