feat(music): prefer vocal tracks, widen the pick's turn budget [spec 03-01] - #123
Merged
Conversation
…03-01] The pick instruction now asks for a song with vocals by default — instrumental-only tracks (light/background music, lofi, piano versions) only when the listener explicitly asks. A real pick under this guidance can spend 5 searches before submitting, so the default turn budget goes from 6 to 8 to keep headroom. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements specs/spec03/03-01-brain-harness.md (music discovery §2.3/§2.5).
What
FIND_MUSIC_INSTRUCTIONnow asks the pick task to choose a song withvocals by default; instrumental-only tracks (light/background music, lofi,
piano versions) only when the listener explicitly asks for instrumental.
DEFAULT_MAX_TURNSfor the pick task goes 6 → 8: a real pick under the newguidance can spend 5 searches before submitting, and 6 turns starved one run
into a null pick (= a silent music break).
Evidence
Two real smoke runs (
murmur-smoke, real Haiku brain + real yt-dlp + ffmpegprobe) on the adversarial scenario — late night, listener asks for "something
soft", the case that used to land on instrumental:
Gates:
pnpm test(761),pnpm run typecheck,oxlint— all green.Peer review (codex gpt-5.5, xhigh): 1 finding, 1 applied (comment carried
edit-history phrasing), 0 dismissed.
AI coding brief
Original request: The listener kept getting instrumental/light music from
the radio's music breaks and wants every pick to be a song with vocals.
Manual interventions: The user called out that the first pass edited the
main checkout directly without the murmur-ship flow; the work was redone
through the full pipeline (worktree, gates, smoke, review). That correction is
now a persistent memory.
Retro: Route any code edit through murmur-ship from the first keystroke —
the "it's just one prompt line" exemption doesn't exist. The smoke run also
paid for itself immediately: it caught the turn-budget starvation a
prompt-only diff would have shipped.
🤖 Generated with Claude Code