القائمة
Use when turning word-level ASR into safe text-based cuts: filler words, 语气词, 口癖, repeated phrases, long silences, caption cleanup, padding, crossfades, and reviewable cut plans.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when the user wants to read, modify, search, add, or delete nodes on a project's canvas. Triggers on "canvas", "scene", "node", "add a clip", "edit the timeline", "what's in the canvas", "rearrange", "remove the X". Use the `clash canvas` CLI subcommands; always pass `--json` for parseable output.
Use when the user wants to list, switch, create, inspect, or delete Clash projects. Triggers on "project", "switch project", "open project", "new project", "delete project", or any reference to a project by name. Use the `clash projects` CLI subcommands.
Use when the user wants to cut, trim, split, reorder, or add media to a Project Timeline, or change its fps or composition size. Round-trip the Timeline through timelines/<id>.timeline.yaml with public `clash timeline list`, `pull`, and `apply` commands.
Use when the user wants to read, modify, search, add, or delete nodes on a project's canvas. Triggers on "canvas", "scene", "node", "add a clip", "edit the timeline", "what's in the canvas", "rearrange", "remove the X". Use the `clash canvas` CLI subcommands; always pass `--json` for parseable output.
Use when the user wants to list, switch, create, inspect, or delete Clash projects. Triggers on "project", "switch project", "open project", "new project", "delete project", or any reference to a project by name. Use the `clash projects` CLI subcommands.
AI video production with the Clash platform. Use this skill whenever the user mentions Clash, video projects, canvas editing, image/video generation, storyboards, or wants to create visual content. Also use when the user asks about managing Clash projects, tokens, or CLI setup.
| name | audio-transcript-cut-planning |
| description | Use when turning word-level ASR into safe text-based cuts: filler words, 语气词, 口癖, repeated phrases, long silences, caption cleanup, padding, crossfades, and reviewable cut plans. |
Use this skill after ASR has produced word-level timestamps. It produces a cut plan, caption projection, and optionally a new non-destructive rendered video asset.
For Clash's current local planner:
clash production plan-text-cut \
--transcript analysis/transcripts/words.json \
--target-asset asset-talk \
--out actions/talking-head-text-cut.json \
--json
Then review the action JSON and run clash production apply-metadata to emit
metadata and caption timeline projections. The caption projection uses
caption items with cue wordIds, source frame ranges, wordRefs, and
sourceToOutputMap. Do not mutate canvas/timeline state without the explicit
CAS apply step. The transcript JSON may include backendId, modelId,
language, durationFrames, and averageConfidence; plan-text-cut records
the transcript file path/hash and those ASR fields in talking-head.analysis.
After review, create a media cut package or ffprobe-validated rendered clean
clip. export-text-cut-media --render refuses to render while the action still
has review cuts; export without --render first when you need a package that
lists pending reviewRanges.
clash production export-text-cut-media \
--action actions/talking-head-text-cut.json \
--source-asset asset-talk \
--output-asset asset-talk-clean \
--out assets/video/asset-talk-clean.mp4 \
--assets assets/manifest.json \
--render \
--json
clash production export-captions \
--timeline projections/timelines/asset-talk.caption.timeline.yaml \
--format vtt \
--out exports/captions/asset-talk-clean.vtt \
--json
clash production project-caption-overlay \
--timeline projections/timelines/asset-talk.caption.timeline.yaml \
--out projections/timelines/asset-talk.caption-overlay.timeline.yaml \
--json
clash production export-timeline-handoff \
--timeline projections/timelines/asset-talk.caption.timeline.yaml \
--format csv \
--out exports/handoff/asset-talk-clean.timeline.csv \
--json
analysis/transcripts/words.json
media/source.mp4
Each word should include id, text, startFrame, and endFrame. Words may
also include confidence and speakerId; transcript-level provenance can
include backend/model/language and average confidence.
actions/talking-head-text-cut.json
projections/metadata/<asset-id>.talking-head.analysis.json
projections/transcripts/<asset-id>.asr-transcript.json
projections/media-cuts/<asset-id>.transcript-cut-plan.json
projections/timelines/<asset-id>.caption.timeline.yaml
projections/timelines/<asset-id>.caption-overlay.timeline.yaml
projections/media-cuts/<output-asset-id>.media-cut.json
projections/media-cuts/<output-asset-id>.ffconcat
assets/video/<output-asset-id>.mp4
exports/captions/<output-asset-id>.srt
exports/captions/<output-asset-id>.vtt
exports/handoff/<output-asset-id>.timeline.csv
reviews/cut-plan-review.md
subtitle-only mode is allowed without video cuts.text clips are allowed for overlays but do not count as a structured
caption system;字幕 projections should be caption items.clash production export-captions when the user needs SRT/VTT/ASS sidecars
for review, upload, or ad delivery.clash production project-caption-overlay when the caption timeline needs
an explicit CAS-required subtitle overlay projection for preview/apply.clash production export-timeline-handoff when a human editor or external
NLE needs a CSV timeline review sheet.Use a controlled reason value:
silencefillertone-particlerepeatfalse-startmanualAggressive mode requires explicit review. Conservative mode may auto-propose but still should not directly overwrite any source asset.