Route, search, inspect¶
SkillRoute gives your agent three MCP tools. Together they turn a vague request into an inspectable shortlist of workflows.
Route a task¶
Find the best skills for adding tests to this Python service. Explain the matches and call out anything ambiguous.
skillroute.route ranks the local catalog against a task. When supplied a repository, it can use filenames and project markers as context. Results include confidence, reasons, source evidence, and clarifying questions when useful. Confidence is a ranking signal, not a guarantee of quality or safety.
Search your library¶
Search my skill library for database migration and schema review workflows.
skillroute.search is useful when you know the subject but want to compare available bundles. It searches the catalog you indexed, not an assumed global collection.
Inspect before using¶
Show the instructions and references for this match before selecting it.
skillroute.inspect_skill makes the matched bundle inspectable. A skill is an instruction set your agent may follow; review its provenance and requested actions before using unfamiliar content.
Explore the Skill Atlas¶
The optional Skill Atlas provides a graph, facets, and matrix views of your library. It is separate from the MCP plugin and uses the Python UI extra. Follow the Skill Atlas guide for setup and screenshots.
Pick the retrieval backend deliberately¶
| Backend | Where the work happens |
|---|---|
| Local token matching | On your computer; default |
| SQLite FTS5 | Local BM25 search |
| Astra DB | Queries your configured remote database when selected |
| LangChain adapter | Depends on the configured integration |
Backend setup explains the optional remote path. Keep database credentials in private configuration, never in skill files or chat.
Understand routing over time¶
Route traces and analytics can help explain why particular skills are selected. Route observability covers stored evidence, timing, and usage views. Evaluation guidance separates measured results from assumptions.