LoRA Learning Skill
Purpose
ruflo adapts small low-rank weights from task outcomes. @ruvector/learning-wasm (shipped with @claude-flow/cli) provides WasmMicroLoRA, WasmScopedLoRA and WasmTrajectoryBuffer. Without it the same code runs on a slower JS fallback.
When to Trigger
- a task succeeded and its approach should be reused
- routing keeps picking the wrong agent or model for this project
When to Skip
- outcomes you have not verified
Commands
Check the backend
npx @claude-flow/cli neural benchmark -i 1000
npx @claude-flow/cli neural status
The MicroLoRA table appears only when the WASM module loads (measured: about 2 µs per 256-dim adaptation).
Train and reuse
npx @claude-flow/cli neural train -p coordination
npx @claude-flow/cli mcp exec -t hooks_intelligence_pattern-search -p '{"query":"auth refactor"}'
Best Practices
- Train only on verified outcomes (tests passed, user accepted)
- Keep one namespace per project
- If the JS fallback is reported, reinstall without
--omit=optional
