Smart Router

Classify a coding task's routing tier — right from VS Code — using the same deterministic classifier that powers the smart-router Claude Code plugin: route a task to the cheapest model tier likely to finish it, and never trust a single model's own word that a fix is correct.

What this extension does

Smart Router: Classify Task — describe a task, and it tells you which tier it would route to (fast / standard / deep), why, and whether it's high-risk or ambiguous enough to need a human in the loop. This works standalone — no Claude Code CLI required.

Smart Router: Install Claude Code Plugin — if you use the Claude Code CLI (in a terminal or via Anthropic's official Claude Code VS Code extension), this one-click sets up the full smart-router plugin: model-tiered subagents, an independent verifier that gates "done" behind a real pass, and a per-repo calibration store that improves routing over time.

Why "classify" and not "route" here

This extension deliberately doesn't try to reimplement the full pipeline inside VS Code. Picking a model tier, running the fix, and independently verifying it only makes sense inside an actual agent session — that's what the Claude Code plugin does. This extension's job is narrower and honest about it: a fast, standalone way to see the routing decision before you commit to it, and a one-click path to the real thing.

Requirements

• For Classify Task: none — it bundles its own classifier, no network or CLI needed.
• For Install Claude Code Plugin: the Claude Code CLI on your PATH.

More

Full design rationale, source, and the underlying Claude Code plugin: https://github.com/aadeina/smart-router

This is v0.1 — the classifier is validated against a 12-task hand-labeled pilot batch (see pilot/RESULTS.md in the main repo), but the full live loop hasn't been measured end-to-end yet. Treat it accordingly.