short-sound-design
Skill4 starsChanged 2 months agoMITView on GitHub
Design and verify audio finishing for Japanese or English short-form social video. Use for 短尺SNS, BGM/効果音, テンポ調整, 意味ベース配置, dialogue-first mixing, or existing rough cut audio finishing with pinned local assets and internal-review QA.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- analyze-footageSkill
- build-blueprintSkill
- compile-timelineSkill
- crop-photosSkill
- design-intentSkill
- evaluate-editSkill
- export-premiereSkill
- finish-business-shortSkill
- finish-creator-shortSkill
- finish-interviewSkill
- full-pipelineSkill
- import-premiereSkill
- re-editSkill
- render-videoSkill
- review-roughcutSkill
- select-clipsSkill
- select-photosSkill
- setup-environmentSkill
- troubleshoot-errorSkill
- vertical-social-compositionSkill
- vertical-social-platform-deliverySkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Start from semantic intent, not an available sound: use a hook accent for the opening promise, an emphasis accent for a meaningful claim or …
- Distinctive: line 21
Keep effects sparse. Leave useful silence, reject effects that merely fill a gap, and inspect repetition across the whole cut for fatigue …
- Distinctive: line 24
Treat dialogue, captions, overlays, lower-thirds, and picture edits as conflicts.
- Distinctive: line 40
Keep these artifacts in the canonical chain: library manifest, promotion result, sound-design request and decision, pinned cue plan, …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Short Sound Design
意味と会話の可読性を先に決め、信頼できるbeat evidenceがある場合だけ小さくtempoへ寄せる。 判断、formal cue、共有音声render、QAを別artifactとして残す。
canonical timeline.json は picture、dialogue、caption timing の正本とし、caption text/timing/approval は
caption runtime と FFmpeg/libass speech route に残す。Remotion/HyperFrames は登録済み content element のみ、
visual treatment patch は別 receipt、AudioRenderPlan/Executor は single mastering とする。rights が未確定の
SFX は fail closed のままにし、agent QA、human audition/approval、platform preview、publication を混同しない。
Practical SFX use
- Start from semantic intent, not an available sound: use a hook accent for the opening promise, an emphasis accent for a meaningful claim or reveal, and a transition accent only when a real visual or editorial change needs support.
- Keep effects sparse. Leave useful silence, reject effects that merely fill a gap, and inspect repetition across the whole cut for fatigue or an accidental pattern.
- Treat dialogue, captions, overlays, lower-thirds, and picture edits as conflicts. Dialogue remains readable; resolve conflict with timing within the evidence window, gain, ducking, or rejection. Never move picture or speech to make room.
- Pin the cue to a semantic window and preserve its source range. Record gain, fade-in, fade-out, dialogue ducking, and tail policy in the cue artifact. Tails may be trimmed to the timeline boundary but may not extend the picture.
A rights or provenance record is a gate, not a suggestion. Missing, unknown, expired, ambiguous, stale, or unreviewed evidence remains HOLD and cannot be selected, rendered, or package-passed. Validate a manifest without media when needed:
npm run sfx:promote -- --asset-id <asset-id> --scope <repo_common|project_local> --manifest <sfx-library.json> --repo-sfx-root <repo-sfx-root> --validate-only --json- Keep these artifacts in the canonical chain: library manifest, promotion result, sound-design request and decision, pinned cue plan, projected timeline, shared audio render plan, mix report, and QA result. A HOLD result is evidence of the missing gate; do not fabricate a cleared entry.
- Machine checks establish identity, hashes, timing, and mix policy. A human must still audition timbre, placement feel, repetition, dialogue masking, gain/ducking, tails, and the final social and delivery renders.
The reusable contracts are command-specific: sound-design:plan accepts only
planning flags; sfx:promote owns promotion flags such as --asset-id,
--scope, --source, --manifest, --repo-sfx-root, --verified-at, and
--permitted-derivatives. Check each command's own --help; do not pass
promotion flags to the plan entrypoint. The machine artifacts are
sfx-library/v1 and sfx-promotion-result/v1; promotion requires a complete
ffprobe/decode measurement and otherwise returns HOLD without creating output.
必ず読む参照
判断前にreferences/workflow.mdを読む。
Workflow
- project、timeline、creative brief、blueprint、source mapのidentityとhashを確認する。 canonical artifactは読み取り専用にし、作業用copyを使う。
- SFX libraryとBGM Packのrights、provenance、manifest、asset hash、sizeを確認する。 未pin、unknown、missing、hash driftは止める。
- picture edit、発言、視覚イベントからsemantic candidateを作る。各候補へpurpose、 evidence、semantic window、asset pinを付ける。
- dialogue、caption、overlay、lower-third、section、music entryをcongestion evidenceにする。 固定間隔の候補生成やpicture timing変更はしない。
solverをdry-runし、人が読めるadopt/reject、score、conflict、snap理由を確認する。
npm run sound-design:plan -- --project <project> --timeline <timeline.json> --request <sound-design-request.json> --decision-output <new-decision.json> --cues-output <new-sfx-cues.json> --repo-sfx-root <repo-sfx-root> --dry-run- decisionを承認候補として確認してから、同じcommandを
--dry-runなしで実行する。 solver decisionからformalsfx-cues/v1だけを生成し、手で別判断へ差し替えない。 cueをA3へ投影する。
npm run sfx:project -- --project <project> --timeline <timeline.json> --cues <sfx-cues.json> --repo-sfx-root <repo-sfx-root> --output <new-timeline.json>Phase 3/4の共有AudioRenderPlan/Executorを使い、social/final audioを同じplanで検証する。 solver内にFFmpeg、mixer、filtergraphを再実装しない。
npm run render-audio-plan -- --project <project> --timeline <timeline.json> --music-cues <music-cues.json> --sfx-cues <sfx-cues.json> --repo-sfx-root <repo-sfx-root> --route social-review --output <new-social-dir> npm run render-audio-plan -- --project <project> --timeline <timeline.json> --music-cues <music-cues.json> --sfx-cues <sfx-cues.json> --repo-sfx-root <repo-sfx-root> --route final --output <new-final-dir>必要なら既存picture、caption、overlayを変えずinternal reviewを作る。
npm run social-review -- --project <project> --timeline <timeline.json> --captions <caption-plan.json> --music-cues <music-cues.json> --sfx-cues <sfx-cues.json> --repo-sfx-root <repo-sfx-root> --output <internal-review.mp4> --work-dir <new-work-dir>- full decode、rational FPS、尺、A/V、plan/report/final-mix parity、LUFS/true peak、 decision/cue/library pin、canonical hash不変を機械検証する。
- 音色・配置・BGM最終選定をhuman audition gate、外部公開をpublication gateとして残す。
禁止
- semantic evidenceなしのhit、固定間隔SFX、BPMだけからの無根拠beat生成
- dialogueを邪魔するhit、未pin素材、低信頼beatへのsnap
- picture timingや発言境界の移動、mixed audioのdialogue finishing再投入
- masteringの重複、内部試写をfinal/public approvalとして扱うこと
- 明示承認なしの公開、upload、外部素材取得
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

