Meta-Skill Routing Simulator¶
When a meta-skill receives a request, it does not load every reference file into context. It inspects the request, extracts keywords, and routes to exactly one guide — loading only what is needed.
This simulator lets you see that decision process in real time.
How it works¶
The meta-skill router follows three steps on every invocation:
-
Keyword extraction. The request is tokenized and stop words are removed. The remaining terms are matched against each route's keyword list.
-
Route selection. The reference file with the most keyword hits is selected. All other guides are skipped entirely.
-
Selective loading. Only the matched guide is injected into context before Claude generates its response.
The routing table¶
| Keywords | Reference file | Library |
|---|---|---|
| chart, bar, line, pie, donut | chartjs-guide.md |
Chart.js |
| timeline, dates, chronological | timeline-guide.md |
vis-timeline |
| network, graph, nodes, edges | vis-network-guide.md |
vis-network |
| map, geographic, location | map-guide.md |
Leaflet |
| flowchart, sequence, diagram | mermaid-guide.md |
Mermaid |
| simulation, physics, animation | p5-guide.md |
p5.js |
| venn, overlap, sets | venn-guide.md |
Venn.js |
Why token efficiency matters¶
Loading all seven reference guides on every invocation would cost roughly 5,600 tokens of context before Claude writes a single line of code. The router loads one guide — typically 650–900 tokens — and skips the rest. That is an 85–90% reduction in reference overhead, which translates directly to faster responses and lower API costs at scale.