What problem does it solve? Agent routing files like AGENTS.md and RESOLVER.md grow to 25-30KB as skills are added, consuming context budget that should go to actual work. This Skill converts hundreds of per-skill routing rows into a dozen functional-area dispatcher entries, cutting file size roughly in half while preserving routing accuracy. ## Core Features & Use Cases - Functional-Area Compression: Groups skills by domain (Brain & Knowledge, Email & Comms, Research, etc.) and replaces N rows per area with one dispatcher entry containing a (dispatcher for: ...) sub-skill list. - Mandatory Verification Gates: Runs gbrain routing-eval --json plus an LLM A/B harness against your edited file before committing, requiring ≥95% lenient routing accuracy. - Eval-Backed Pattern: Ships with A/B results across Opus, Sonnet, and Haiku showing the dispatcher pattern beats both the 270-row baseline and resolver-of-resolvers compression. - Use Case: Your AGENTS.md hit 30KB and context-health checks flag it red. Run this Skill to compress it into 13 functional-area entries, verify routing accuracy with the bundled harness, and commit only after both gates pass. ## Quick Start Ask the agent to compress your oversized AGENTS.md or RESOLVER.md routing file into functional-area dispatchers and verify routing accuracy before committing.
