ratel-tune-definitions
/ratel-tune-definitions — make tool & skill definitions retrievable and usable
Ratel retrieves tools and skills over their definitions — a lexical (BM25) index rewards descriptive names, parameter names, enum values, and exact trigger terms, while, in semantic/hybrid mode, a semantic index matches the description's meaning — and the model then decides which of the top-K hits to call. Both steps fail on bad definitions: a vague description surfaces in neither arm; a bloated one drowns its neighbors; two near-duplicate tools split the model's confidence; a loose schema lets the model call a tool wrong. This skill is the manual tuning pass that fixes those — definition by definition, with concrete before/after rewrites.
The deliverable is a markdown plan at <repo>/.ratel/ratel-tune-definitions.md: a per-definition before/after table and a prioritized fix list. The plan is implementable by the customer; this skill does not edit their tool or skill code in place.
This skill is the fix that the /ratel-assessment "Definition Quality" dimension routes to. It pairs with:
/ratel-decompose-prompt— that skill extracts skills from a prompt; this one sharpens theirdescriptionandtagsso they're actually retrievable. Run this after a decomposition on the extracted set./ratel-assessment— the front-door audit that flags weak definitions and points here.
LLM-driven suggestions — automated description rewrites, missing-parameter detection, and redundant-tool merge proposals — are the direction of Ratel Cloud's suggestions engine (Coming Soon), which derives ranked suggestions server-side from your exported telemetry. This skill is the manual version of that pass; note in the plan which fixes the customer could defer to that engine once it ships.
Philosophy
Three rules. Break any of them and you've either made the catalog prettier without making it more retrievable, or you've optimized for retrieval at the cost of the model getting confused.