메뉴
Use when working with audio features, BPM, Camelot keys, transition scoring, GA set builder, iCloud stubs, M3U8 export, or cheat_sheet generation. Triggers on app/utils/audio/, TrackFeatures, scoring, deliver_set, cheat_sheet.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when building DJ sets, scoring transitions, delivering sets, syncing to YM playlists, or improving set quality. Triggers on build_set, deliver_set, score_transitions, rebuild_set, "построй сет", "экспортируй сет".
Use when delegating tasks to Codegen cloud agents, monitoring agent runs, reviewing agent PRs, or running multiple agents in parallel. Triggers on "delegate", "codegen", /delegate, "@codegen-sh", Codegen Bridge.
Use when adding a new MCP tool, registering tools in gateway, creating Pydantic types for structured_content, or writing MCP tool tests. Triggers on app/mcp/tools/, FastMCP, register_tools, "new tool", "add instrument".
Use when expanding playlists with new tracks, discovering similar tracks, importing from YM, auditing quality, or distributing by subgenre. Triggers on "расширь плейлист", "найди похожие", "заполни плейлист", "expand playlist", "fill playlist", discover_candidates, audit_playlist, distribute_to_subgenres.
| name | audio-analysis |
| description | Use when working with audio features, BPM, Camelot keys, transition scoring, GA set builder, iCloud stubs, M3U8 export, or cheat_sheet generation. Triggers on app/utils/audio/, TrackFeatures, scoring, deliver_set, cheat_sheet. |
Audio анализ — дорогая операция. Если фичи уже есть в DB — используй их. dj_analyze_track требует dj_activate_heavy_mode.
| Запрос | Инструмент |
|---|---|
| BPM / key / energy трека | dj_get_track + dj_get_features (DB, мгновенно) |
| Посчитать фичи | dj_analyze_track (ML, 15-120 сек, тег heavy) |
| Треки по параметрам | dj_filter_tracks (BPM, key, energy) |
| Построить сет | dj_build_set (использует готовые фичи) |
| Оценить переходы | dj_score_transitions |
| Экспортировать | dj_deliver_set (score → файлы → опционально YM) |
| Компонент | Вес | Hard constraint |
|---|---|---|
| BPM | 0.30 | diff > 10 → 0.0 |
| Harmonic | 0.25 | Camelot dist ≥ 5 → 0.0 |
| Energy | 0.20 | LUFS diff > 6 → 0.0 |
| Spectral | 0.15 | — |
| Groove | 0.10 | — |
Пороги: ≥ 0.85 хороший, < 0.85 слабый (!!!), 0.0 жёсткий конфликт.
TransitionScoreResult для каждой пары. Hard conflicts → вопрос пользователю.generated-sets/{name}/def is_local(path: Path) -> bool:
st = path.stat()
return st.st_blocks * 512 >= st.st_size * 0.9
# shutil.copy2 на стабе = TimeoutError!
Пропускай стабы при экспорте, в M3U указывай путь к исходному файлу.
@dataclass(frozen=True, slots=True)
class TrackFeatures:
bpm: float; energy_lufs: float; key_code: int # 0-23
harmonic_density: float; centroid_hz: float
band_ratios: list[float]; onset_rate: float
mfcc_vector: list[float] | None = None
kick_prominence: float = 0.5; hnr_db: float = 0.0
Конверсия ORM→TrackFeatures: app/utils/audio/feature_conversion.py.
dj_analyze_track без явного запроса — занимает минутыshutil.copy2 — TimeoutErroravg_score ≥ 0.75Подробности: .claude/rules/audio.md.
NO FEATURE INTERPRETATION WITHOUT CHECKING COLUMN NAMES IN DB SCHEMA
onset_rate vs onset_rate_mean, hnr_db vs hnr_mean_db — неправильное имя колонки = NULL = сломанный scoring. Перед ЛЮБЫМ SQL или ORM запросом к audio features — проверь .claude/rules/db-schema.md.
| Отговорка | Реальность |
|---|---|
| "Я помню имя колонки" | onset_rate vs onset_rate_mean уже ломало запросы дважды |
| "hp_ratio в диапазоне 0-1" | hp_ratio UNBOUNDED (0.66-17.25), НЕ нормализован |
| "Запущу анализ для всех треков" | Анализ = 15-120 сек × N треков, нужен heavy_mode + согласие юзера |
| "Фичи одинаковые для всех pipeline" | v1.0 не имеет beats (kick, onset, pulse = NULL), только v2.1b6 полный |