Development Log: Second Brain Trace Quality Tuning

Purpose

Improve Second Brain Trace quality after a real answer showed noisy raw conversation notes, misleading "used" wording, and overly dramatic answer style.

Implementation Summary

Excluded raw conversation and transcript folders from default trace retrieval unless the user explicitly asks for raw/transcript content.
Penalized index files so curated notes are preferred.
Boosted curated record, decision, planning, and development paths.
Changed visible wording from "used notes" to "selected notes" / "selected for answer context" to avoid overstating actual post-answer usage.
Added trace context guidance telling the model not to imitate dramatic wording from retrieved notes.
Added base prompt guidance against phrases such as final execution mandate, engineering standard, knowledge distiller, and Antigravity's yardstick unless the user explicitly asks for that style.
Expanded tests to verify curated notes beat raw conversations and index files.

Changed Files

src/features/secondBrainTrace.ts
src/utils.ts
tests/secondBrainTrace.test.ts

Verification

./node_modules/.bin/tsc --noEmit
npm run compile
./node_modules/.bin/jest --runInBand

Result

Second Brain Trace should now surface more useful curated project notes, be clearer about what was selected as answer context, and reduce style contamination from raw notes.

1.4 KiB Raw Blame History