video-editing
Skill0 starsChanged 41 days agoMITView on GitHub
Edit spoken videos, shorten pauses with word-timed transcripts, compare indoor/outdoor grades, normalize audio, review revisions, and export flat Final Cut Pro XML using Video Edit Harness. Use for footage editing rather than screen recording or synthetic narration alone.
58 other projects use a version of this.
Adopted by
Most recently taken up by
- AdeyStudios/skillstheir own version, first seen Nov 16, 2026
- keny4nn/claude-skillstheir own version, first seen Nov 16, 2026
- trku052-png/social-media-youtube-skillstheir own version, first seen Nov 15, 2026
- notque/vexjoy-agenttheir own version, first seen Nov 10, 2026
- Rojios/everything-claude-codetheir own version, first seen Nov 8, 2026
- ourostack/ouroboros-skillstheir own version, first seen Nov 8, 2026
Used with
Projects with it also use everything-claude-code-conventions, agent-sort and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
mryeastcodeSkilleverything-claude-code-conventions
Development conventions and patterns for everything-claude-code. JavaScript project with conventional commits.
Also in 8 projects that use this
niloykumarbarmanSkillagent-sort
Build an evidence-backed ECC install plan for a specific repo by sorting skills, commands, rules, hooks, and extras into DAILY vs LIBRARY buckets using parallel repo-aware review passes. Use when ECC should be trimmed to what a project actually needs instead of loading the full bundle.
Also in 8 projects that use this
InfopibeSkilleverything-claude-code
Also in 7 projects that use this
tpavanipradeepSkillnextjs-turbopack
Next.js 16+ and Turbopack — incremental bundling, FS caching, dev speed, and when to use Turbopack vs webpack.
Also in 14 projects that use this
tpavanipradeepSkillbun-runtime
Bun as runtime, package manager, bundler, and test runner. When to choose Bun vs Node, migration notes, and Vercel support.
Also in 14 projects that use this
niloykumarbarmanSkillproduct-capability
Translate PRD intent, roadmap asks, or product discussions into an implementation-ready capability plan that exposes constraints, invariants, interfaces, and unresolved decisions before multi-service work starts. Use when the user needs an ECC-native PRD-to-SRS lane instead of vague planning prose.
Also in 8 projects that use this
- +2 more from tpavanipradeep
- +1 more from niloykumarbarman
Other versions58 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 58
- affaan-m/ECC★ 246ksame name
- ZJU-REAL/Easel★ 4.7ksame name
- notque/claude-code-toolkit★ 450same name
- notque/vexjoy-agent★ 450same name
- loulanyue/awesome-claude-notes★ 272same name
- maxazure/video-editing-skill★ 204same name
- tpavanipradeep/everything-claude-code★ 104same name
- mturac/everything-openai-codex★ 89same name
- mit-network/everything-claude-code★ 76same name
- Fmarzochi/EGC★ 61same name
48 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 0 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 52 edited versions we read keep these.
0
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Run session context SESSION --words.
“Common” is measured: kept by at least 6 in 10 of the 52 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video editing with Video Edit Harness
This skill requires the Video Edit Harness checkout (release v0.1.0-alpha.7), Python 3.11+, uv, FFmpeg and ffprobe. The standalone skill ZIP contains instructions and its license; it does not bundle the harness or media.
Find the harness checkout. If VIDEO_EDIT_HARNESS_ROOT is set, use that directory. Otherwise, find the checkout containing pyproject.toml, video_harness/, and this skill; if the skill was copied elsewhere, locate the separately prepared checkout or follow the installation guide. Ask for its location only when it cannot be determined; do not install dependencies without a request. Run commands from the harness root with uv run video-harness. Run uv run video-harness doctor first: check the reported root, versions, skills and commands; resolve relevant findings before editing. Read its README.md and AGENTS.md; use uv run video-harness session --help for exact command syntax.
Use the current request and existing authorization to establish scope before any cloud upload, message, or publication; ask only when that scope remains unresolved. Honor explicit no-upload and no-publication restrictions for the actual source, including private footage of children. Publishing this skill or harness never authorizes sending its users' footage. With protected footage, continue local grading/audio/technical checks and use an existing sealed transcript when available; if there is none, do not cloud-transcribe it or invent word timings. Use an explicitly permitted alternative source for public demos. Cloud transcription needs authorization for the particular source and credentials in environment variables. The route is OpenAI then Azure OpenAI; there is no local Whisper/ASR. Never synthesize human approval or observations. Record your real actor and keep source footage intact.
For color-only requests, use preview PROJECT --styles natural,warm_documentary,soft_film (or styles matching the brief), compare the same source interval, and record the user's choice. A transcript and editorial session are required only for transcript-backed story/pacing edits; do not upload a source just to grade it. Read docs/PRESET_CATALOG.md in the harness for use-case/style selection.
- Copy
projects/talk.template.jsonfor the project. Replacesourcewith the actual absolute media path, setinput_colorand preview interval, and check media tracks. Startsession start PROJECT SESSION --brief-file BRIEF --actor ACTOR, then inspectsession status SESSION --deep. - Persist the existing source-specific decision once:
session cloud-policy SESSION allow --providers openai azure --actor ACTOR --note BASIS, ordenywhen uploads are forbidden. The policy is tied to the source SHA; resuming never turns deny/unknown into allow. Do not ask again for an unchanged source/provider already authorized. If source upload is authorized, useuv run --extra transcription-cloud video-harness session transcribe SESSION. Or attach a sealed existing transcript withsession transcript SESSION TRANSCRIPT_JSON. Read the transcript, words, and mismatch warnings. Correct spelling with exact word IDs throughsession correct-transcript; never change timestamps by guessing. - Run
session context SESSION --words. Build a story spec with ordered chapter spans anchored to observed word IDs and omissions with exact IDs, reasons, and goal IDs.examples/workflow-story.jsonshows structure only. Runsession plan SESSION --spec-file SPEC_JSON --actor ACTOR. Inspect the plan, thensession approve SESSION --actor ACTOR --note REASON. Agent selection is recorded as agent selection. - Render a preview with
session render SESSION. Review video, audio-only output, subtitles, and frame mapping. Usesession inspect SESSION RENDER_ID --start SECONDS --duration SECONDSfor focused filmstrip, waveform, listening audio, and source words. Usesession audition-junctions SESSION RENDER_ID --offset 0 --limit 24and subsequent offsets or--idsto inspect every needed boundary. The gallery shows remaining IDs. Listen across cut boundaries; protect word endings, breath, emphasis, and topic changes. Visual aids alone are not a listening review. - Create a feedback page with
session review-page SESSION RENDER_IDand import its JSON withsession feedback SESSION --data-file FILE. Usesession revisefor edit operations orsession address-feedbackfor other changes. Re-render and review candidate fixes. Record a six-check review (meaning, pacing, audio_only, cut_boundaries, captions, color) bound to the precise render SHA-256. For more than 24 boundaries, passing cut review also requires exact-SHAjunction_coverage:full_listeningwith a note, orselectedwithreviewed_idsand a reason for eachwaivedID. Do not label generated audition files as listened to. Example review JSON is an unreviewed shape. - After acceptance, run
session render SESSION --full; review that exact full render, thensession package SESSION RENDER_ID --target fcp. The XML supports a flat, single-source timeline. In actual FCP on macOS, check import, playback, grade/LUT, captions, and mix. Record only performed checks withsession delivery-check; usesession finishwhen its gates pass. The sealedcompletion.jsontravels inside the delivery folder;delivery.jsonremains the original packaging snapshot. A synthetic completion proves only the fixture workflow, not human acceptance. LUT, SRT, spatial masks, and final audio may need separate application.
Use session handoff and session resume when switching agents. Keep run logs and hashes. Distinguish synthetic/technical checks, human visual and listening review, FCP GUI proof, and distribution playback. For color comparisons, keep source interval and audio conditions fixed; adopt only a human-reviewed look. Apple Log LUTs already transform to Rec.709, so avoid a second FCP Camera LUT. See README.md and docs/PRESET_CATALOG.md.
In alpha.7, new visual sessions use visual_pipeline_version: 2: grade assembled footage before overlays and effects, retaining visual-graded.mp4. Do not apply the footage LUT to completed UI or baked FCP picture. Existing unversioned projects preserve pipeline 1; migrate through an unadopted candidate, inspect visual-pipeline.json, and retrack the actual pre-effects stage after a grade/pipeline change. Secondary image/comparison colors remain their own; deliberate picture effects can still affect existing overlays. See docs/VIDEO_EFFECTS.ja.md.
Optional editing patterns: shared contract
For development external finishing returns, read docs/TIKTOK_API.ja.md. native-inspect checks a downloaded MP4 locally; request --require-music only when music was requested. session native-result retains returned bytes against an existing proposal and an actual source/output SHA receipt. Reported UI/API execution is a declaration, not verified execution. Absent or near-silent audio does not satisfy requested music; audible audio does not identify the track. Preserve the retained file and raw logs; review the exact returned SHA. Do not inherit source mapping, captions, beats, tracking or approval after an external edit, or bind unrelated stock footage to a local source. This route neither operates the external editor nor posts, adopts or establishes cross-platform rights. It is not in alpha.7.
Use docs/TRACKING.ja.md for declared composition protection and local manually seeded tracking. composition_guides protects fixed regions or explicit source-bound tracked boxes against supported title/zoom conflicts. Choose LK for textured features, CSRT for appearance tracking, or optional MediaPipe pose for an observed torso; inspect the actual results rather than assuming a backend works for every dance. session tracking-source identifies the retained pre-effects picture; session track-effect tracks an observed frame interval and proposes an unadopted tracked_zoom candidate. Track the exact picture stage used by the renderer. Lost frames, changed input or tracking SHA, and overlapping zooms must fail rather than inventing positions. Pose array indices and geometric/appearance scores are not person identities. Manual correction requires observed boxes. session track-effect adds a tracked subject guide: every valid frame is checked for measured title collision and zoom crop. Tracked guides reject split/comparison geometry. Torso boxes do not protect unobserved faces or feet; Automatic title relocation is not implemented. When the current effects-catalog includes tracked_title, use session track-effect --effect tracked_title --title-parameters-file for an explicitly chosen label. Select above/below/left/right and inspect every frame; clipping, lost/stale tracking, declared-region overlap and overlapping zoom/split/comparison geometry must fail. Use docs/RETIME.ja.md for source-bound standalone ramp/hold rendering with explicit protected source-frame intervals and observed captions. Prefer retiming source/environment audio before placing added BGM on the new output timeline. Rubber Band R3 is the default audio backend; the phase-vocoder alternative must be explicit and auditioned. Freezes insert silence. The renderer remaps captions to an SRT and reports skipped captions; it does not invent word timing. Visual-session retime-source / retime proposes unadopted timing candidates with source-frame provenance. Alpha.7 invalidates old explicit cues/effects/guides; development migration follows the retime guide. Added BGM is placed after retiming. Baked mix/video_only FCP handoff retains the result; editable FCP retime and tracking migration remain pending. Speech word protection uses the nonspoken-interval workflow below. Regenerate tracking after a timing change and rebuild cues when explicitly cleared; technical frame/sample checks do not establish human listening quality.
For a request to make the edit less monotonous, first use the actual source to shape a hook, a clear promise, rising information or emotion, and a payoff. Vary shot scale and duration at changes in meaning; choose B-roll that shows the named action, place, object or evidence rather than generic motion. Use a motivated J-cut or L-cut when adjacent source audio and picture permit it, and keep the speaker's words and meaningful ambience intelligible. A selective punch-in, sound hit, brief contrast change or color accent can mark one important turn; leave quieter space before or after it. Keep captions readable and clear of faces, hands, demonstrations and platform UI. Do not apply an effect at every cut or claim it will improve watch time.
Read docs/VIDEO_EFFECTS.ja.md before promising a requested visual effect. Translate the requested type, intensity and timing into a video_effects project setting: either a pop_dance/subtle preset with low/medium/high intensity, or explicit version-1 events bound to the exact current frame-mapping SHA. Supported event types are zoom_pulse, split_screen, monochrome and color_frame; each explicit event needs an ID, output start/end, strength from 0 to 1 and a concrete reason. A preset anchors to observed cut boundaries, not imagined beats. For a beat-timed effect, use only an actual mapped beat position from the selected local music and its current cue plan; an unknown beat phase remains pending. natural or an explicit off direction disables additions; contradictory effect settings must fail visibly. Unsupported speed ramps, glitches and FCP third-party plugins must be reported as unsupported rather than counted as applied.
For comparison_wipe, bind a separate registered video with asset_id, source-frame-aligned source_start, and layout: wipe or side_by_side. Wipe assumes comparable framing; side-by-side retains both pictures with padding. Preserve base audio only; validate secondary rights, SHA and source range. For keyword_title, use explicitly chosen text, measured bounds, readable palette, and fade/rise motion. Record font and generated PNG hashes. These helpers do not detect subjects or prove reading time; inspect faces, hands, existing captions and platform UI.
For parameterized local effects, inspect video-harness effects-catalog. smooth_zoom supports manual normalized anchors, scale and easing; saturation_pulse supports saturation and easing. These are not automatic subject tracking. Use session effects SESSION RENDER_ID --operations-file FILE --actor ACTOR --note REASON to add, update or remove events as an unadopted candidate based on that render. Preserve the quiet version and inspect entry, peak and exit. Comparison pages synchronize playback and export selection proposals; a downloaded selection file is not final approval. See docs/VIDEO_EFFECTS.ja.md for operations.
Render the plan and inspect effect evidence plus the actual frames across every requested event, including entry, peak and exit. Compare with the quiet version over the same source coverage and audio conditions. Record requested versus applied events, reasons for any rejection, the exact output SHA and a real human viewing/listening decision. A technical render is not perceptual approval. Finished-picture mix or video-only handoff can burn effects into the MP4; do not promise these effects as editable FCP layers. If editable handoff rejects the plan, choose an authorized supported mode or leave the handoff unresolved. The effect implementation and its tests must be confirmed in the current checkout before calling this path supported.
Read docs/EDITING_PATTERNS_USAGE.ja.md for the command contract. session propose-beats SESSION CURRENT_RENDER_ID --spec-file SPEC --actor ACTOR --note REASON binds a music beat map to the exact render. Explicit nonspoken intervals are required for visual cut movement; the proposed plan needs selection and a new render. Speech plans receive marks only. Never infer permission to cut speech from a low-amplitude signal. session import-production-fcp SESSION CURRENT_FULL_RENDER_ID XML --actor ACTOR --note REASON retains a returned production XML and a candidate changes file for supported editable cues; use candidate, render, comparison and explicit adoption. Neither import nor adoption grants human review. Pexels search/fetch have separate permission gates, and a download receipt does not establish rights approval.
For a checkout that exposes video-harness production --help, resolve creative direction separately from color (style) and scene (use_case). Default to natural, with no added music, effects, stock visuals or beat sync. A platform name or “trendy” alone does not choose busy editing. The alternatives are gentle_vlog, clear_explainer, cinematic_story, beat_montage and playful_short; respect explicit per-effect off settings. Save a lasting preference only on an explicit user request, recording the actual actor and reason.
Use production register SOURCE --metadata METADATA_JSON --output ASSET_JSON for a local asset. Preserve its license evidence, source URL, credit and SHA. Check production validate ASSET_JSON --policy POLICY_JSON for the actual outlets and use. Free price does not establish commercial, advertising, Content ID or redistribution rights. embedded_use, mixed_audio_handoff and raw_asset_handoff are independent. Unknown conditions remain unresolved. Prefer the user's specified asset and own footage, then eligible project/local assets; external search and download have separate permission gates and must not send private transcript text automatically.
For an already selected edit plan, create a comparison with session candidate SESSION --changes-file CHANGES_JSON --actor ACTOR --note REASON, render through session render SESSION --candidate-id ID, and use session compare-candidates SESSION RENDER_ID RENDER_ID. Keep the no-addition option. session adopt-candidate SESSION ID --actor ACTOR --note REASON selects the exact immutable snapshot; it never grants human approval or a full-render review. Pattern, music, cue or asset changes require a new exact-render review and matching captions/portrait derivatives.
Explicit cue plans bind to the rendered mapping SHA; use production cues PROJECT_JSON FRAME_MAPPING_JSON --output PRODUCTION_JSON to inspect proposed positions. Listen for voice clarity, pumping, loop seams and endings. Beat analysis (production beats) is local music signal analysis, not ASR. Inspect ambiguity and frame rounding; do not cut or retime words to follow a beat. Mapping beats does not itself adopt a revised edit.
For footage without speech or an audio track, start a project with edit_basis: visual and actual registered source assets. Use session visual-plan SESSION --plan-file VISUAL_EDL_JSON, inspect the source-frame selections, then explicitly select the plan. Do not invent a transcript or word IDs. Source environment audio is preserved; missing audio is explicit silence. Review visual context and cuts, adding listening checks when source sound or BGM is present.
Read the local implementation status before promising FCP production handoff. XML/DTD success is not GUI import, playback or re-export proof. Finished-picture mix mode already carries its grade; never apply its conversion LUT again. Editable layers may still require ducking, grade and overlay finishing. Package only the licensed mode, and report every pending observation rather than claiming a completed FCP round trip.
For a catalog advertising title supported_versions [1, 2], version-2 keyword_title/tracked_title can wrap Japanese or English using the local text-layout extra. Set language, max_width_fraction, max_lines and protected_phrases explicitly. Preserve named products and units; never truncate or shrink automatically after overflow. Engine/model bindings and measured wrapped bounds must survive rendering and collision checks. Inspect actual lines and reading time on a phone; wrapping is not semantic or perceptual approval. This capability is included in alpha.7.
For wrapped tracking labels, session track-effect accepts --title-version 2 with the title parameter JSON. Preflight uses the retained pre-effects frame size before tracking, returns measured lines/bounds and binds font/layout evidence; a successful fit does not prove tracked placement is clear of subjects or UI. Keep the candidate unadopted until its rendered comparison is reviewed.
In a checkout with title_collision_checks, rendering rejects overlapping measured keyword/tracked cards on shared visible frames, even without composition guides. Resolve the reported IDs/frame by revising position, width or timing as an unadopted candidate; do not silently move or omit the requested text. Guide-free title clearance proves only title-to-title geometry, not subject/UI clearance or sufficient reading time. Existing overlapping plans may now need revision before rerendering.
In alpha.7, a newly rendered speech session retains speech-base.mov before production additions and normalization; session retime-source returns word_protection bound to the exact plan/transcript/mapping and source. Use original word IDs and occurrence indices, not rounded subtitle times, to inspect protected frames. Unlisted audio is not automatically nonspoken. Alpha.7 speech session retime accepts source-bound version-2 settings with explicitly observed nonspoken_intervals, mandatory word protection and optional additional protection. Inspect/listen before recording observations; never infer nonspoken intervals from missing words or low amplitude. Render the unadopted candidate to remap grouped subtitles and exact original-frame correspondence, retaining full-resolution assembly and retimed PCM before added audio and normalization. Old explicit cues/effects/guides are invalidated; reapply on the new mapping. Keep original-cut-reference XML and use baked mix/video_only handoff; editable FCP retime remains unsupported. Review the exact final MP4 before adoption or delivery.
For speed/hold comparisons, use session compare-candidates SESSION NATURAL_RENDER RETIMED_RENDER --mode timing for independent full-length playback with the same selected plan/source coverage/color/base audio/preview conditions. Read duration and retime-operation evidence; do not treat equal output seconds as equal source moments. The default effects mode remains synchronized and strictly mapping-bound. Timing mode does not compare reordered or different plans. A page selection is a SHA-bound proposal, never adoption or final approval.
For structure alternatives, use session compare-candidates SESSION FIRST_RENDER SECOND_RENDER --mode structure with different sealed plans under the same brief/source pool/color/base audio/FPS/preview conditions. Include a natural version without retiming. Inspect ordered original source ranges and reasons, added/removed ranges including repeated uses, duration and retime operations. Play each full video independently; equal output times do not mean equal source moments. Selection exports are exact-video-SHA proposals, not adoption or human review. The default effects mode and same-plan timing mode retain their stricter boundaries. See docs/RETIME.ja.md; this capability is included in alpha.7.
Import a downloaded comparison proposal with session select-comparison SESSION --data-file comparison-selection.json --actor ACTOR --note REASON. The unadopted candidate restores the selected render's exact plan/project and retained transcript/context inputs, not the current plan with old effects. It validates video SHA, current brief, source and output time. Render the candidate before explicit adoption, then review the exact full output. Restoring older inputs does not restore human approval or publish a video. Legacy speech renders without input references require their actual embedded transcript; context is regenerated from that transcript and the same brief rather than inventing alignment. Included in alpha.7.
For J/L cuts, read docs/AUDIO_CUTS.ja.md and use session audio-cuts on an exact existing render. Declare actual source-handle and replacement listening observations, handle_audio_kind, complete transcript word IDs and repeated-word acknowledgment. Never infer nonspoken audio from missing words. Untranscribed visual speech is unsupported. Render the unadopted candidate, inspect separate audio provenance and rebuilt captions, and require an exact-SHA listening review. Use baked mix/video_only; retime combinations and editable FCP are unsupported. This feature is included in alpha.7.
For motion_trail, read docs/VIDEO_EFFECTS.ja.md and propose an explicit session effects add operation on the actual mapping. Limit it to an observed movement within one shot; do not add it to natural edits. Render the unadopted candidate and compare against natural picture with the same audio. Inspect duplicated faces/hands, baked text, low-FPS stepping and phone readability; temporal_inputs proves frame/weight provenance, not aesthetic acceptance. It is full-frame temporal mixing, not a subject mask. This feature is included in alpha.7.
For experimental tracked_background, read docs/TRACKING.ja.md. Use a full-subject observed box on the retained pre-effects stage, not a torso-only pose box; pose requires manual masks for every selected frame. session track-effect --effect tracked_background prepares local GrabCut masks as an unadopted candidate. Keep the manifest/PNG/correction hashes and inspect every contour, hair, hands, clothing and occlusion on the exact rendered comparison. Manual full-size binary PNGs replace individual masks; never hide loss or fall back to a rectangle. Require the SHA-bound subject_masks visual check. Baked mix/video_only only; no depth or semantic matting claim. This alpha.7 feature is experimental: an eight-frame real-footage trial kept wall/ceiling in the foreground and the agent rejected the visual result. No human review is recorded. Require full contour review and manual masks before use; do not claim universal person matting.
The post-alpha.7 development checkout offers optional semantic background masks through session track-effect --effect tracked_background --algorithm pose --model LOCAL_TASK --mask-backend pose. Confirm these flags in the installed CLI before using them; alpha.7 retains the GrabCut/manual-mask path. Seed the observed torso, retain source/model/library/threshold bindings, and stop on identity loss or ambiguous/mismatched poses. Use full-size binary PNG corrections for incomplete outlines; require the exact candidate render's subject_masks visual review. A mask can still lose hands or include another person's overlapping limbs. Do not adopt or call the final MP4 complete from model confidence, successful decode or the agent's frame inspection alone. See docs/TRACKING.ja.md.
For the development session motion-cuts path, derive adjacent endpoint choices and subject ROIs from actual source frames. Declare only observed nonspoken frame intervals, inspect measured motion and uncertainty, and require explicit plan selection before rendering. Preserve the source/hash/frame/FPS report and rebuild invalidated cues/effects/guides against the revised mapping. Compare structure changes as structure, retain the original render, and obtain new exact-render visual/listening review. Optical direction compatibility does not establish action continuity or camera-motion compensation; this session command does not render overlapping transitions. For explicitly requested overlapping transitions in a development visual session, use session transitions on the retained assembly and read docs/TRANSITIONS.ja.md. It creates an unadopted source-bound dissolve/push candidate and clears old cues/effects/guides. Match preview/full conditions, inspect both actual source references and separate unchanged base audio, compare with natural, and require the exact final video SHA's transitions visual check. Natural/off, retime/JL combinations, Apple Log/HDR and editable FCP transitions are unsupported; use baked mix/video_only. Successful synthetic tests do not establish real-media, human or FCP GUI acceptance. See docs/MOTION_CUTS.ja.md.
For development compound effects, read docs/MOTION_TEMPLATES.ja.md and inspect motion-templates. Use session motion-template on an actual render with explicit frame-aligned interval, observed purpose, parameters and reduced_motion choice. The beat_focus recipe combines zoom/saturation; reveal_callout sequences zoom and a supplied keyword. Reduced variants limit animation, not all visual change. No beat, depth or native Motion/TikTok analysis is implied. Recipes remain unadopted and retain source-render/mapping SHA provenance; compare a natural baseline and review the exact final video. Do not bypass natural/off, composition or text collision checks. This development feature is not in alpha.7.
For development relative-depth image layers, read docs/DEPTH_LAYERS.ja.md. Prepare or infer fields on the full render’s retained visual-graded.mp4, preserving raw fields and interval-shared normalization; use depth correct for observed frame corrections. If flicker warrants it, propose opt-in depth stabilize with declared source-frame cuts; preserve the original, inspect recorded coverage/resets and compare actual moving contours. Low confidence resets history. Propose session depth-layer with a registered same-canvas RGBA asset and permitted visual-asset policy, compare a natural baseline, and retain the candidate through selection/adoption. Require exact-final-SHA depth_contours visual review of hair, hands, all performers, occlusions and flicker. Unrelated audio/effect revisions retain depth; changed grading must regenerate fields. Natural reset removes depth. Baked mix/video_only handoff preserves redacted field provenance; models and original images remain external. Relative depth is not a person mask, metric distance or guaranteed temporal consistency. A technical pass is not human acceptance or a finished video. This development feature is not in alpha.7.
For development portrait caption wrapping, read docs/CAPTION_LAYOUT.ja.md. Use an explicit ja/en layout and observed protected terms with tiktok-export --caption-layout; preserve reviewed SRT timing and manual breaks. The actual font/stroke ink must fit three measured lines; revise oversized cues rather than inventing timing or splitting terms. Inspect missing glyphs, reading time, subjects and platform UI on the exact portrait render. Portable delivery retains visible-line/runtime proof without input paths or unused protected terms. This is not semantic transcript regrouping, complete JLREQ compliance or an official platform safe zone; it is not in alpha.7.
For development editable FCP visual cues, read docs/FCP_OVERLAY_PLACEMENT.ja.md. Image/video/title exports retain static placement and actual source dimensions through a shared preview pixel model; preserve its pending FCP GUI calibration status. Inspect geometry against the exact preview, and do not claim playback parity from DTD or XML comparison alone. Return import accepts unchanged placement and rejects unknown spatial edits/locations/automation. Audio ducking/normalization parity remains a separate unfinished stage. This is not in alpha.7.
For development audio-cue renders, read docs/AUDIO_GAIN_EVIDENCE.ja.md. Retain measured sample gain arrays and the actual speech/environment and mixed WAV inputs; cache restoration must validate their fingerprints. For an explicitly requested experimental editable handoff, set fcp_handoff: editable and fcp_audio_automation: measured; bounded dB keyframes and separate dialogue PCM retain pre-normalization gains. Apply a final common normalization scalar only after all-sample proof against retained pre-AAC WAVs. Return import allows supported visual cue revisions while preserving all generated audio; reject changed audio automation/ranges/dialogue. Preserve pending FCP GUI calibration; inspect original-audio muting, loop source clocks and interpolation in FCP. Final normalization/playback parity remains pending. Local evidence references can contain private paths. Review the exact MP4 by listening; never infer human approval from gain data. This is not in alpha.7.
For development visual retime, read docs/RETIME.ja.md: default --timeline-settings migrate maps compatible explicit cues, pulse effects and fixed guides through exact frame membership into an unadopted candidate. Preserve the sealed migration evidence and inspect the actual rendered mapping. Phase maps retain old-frame zoom/saturation values and keyword-title fade/rise during holds and omissions. A title beginning at a surviving visible old frame must still pass collision checks. BGM plays normally on the new clock; its declared beat offset follows the music. Video overlays use version-2 phase maps selecting actual-output RGBA samples sealed by the original render, rather than reconstructing input clocks. Preserve the layer fingerprint and source/trim/placement digest. Changed opacity/placement/trim, layer tamper or preview canvas mismatch requires a fresh full original render; do not upscale the layer. Old timestamp-only maps need a fresh original render. Private layer references stay in the local session; baked MP4 retains the picture. Sealed version-2 video layers preserve loop position and baked fade. SFX content, including loops, follows the continuous picture map with pitch-preserving audio retime; use baked mix/video_only handoff. Wipe/trail content uses sealed original-stage samples through content_map when the original render retained them. Keep original event/plan/asset/order and picture-setting bindings, original and remapped layer hashes, and exact frame selections. Changed grade, overlays or effect parameters requires a fresh original candidate. Old renders without samples allow only complete one-to-one translation; speed changes need a fresh full render. All temporal stages keep their captured order when omissions collapse start times. Stale/omitted/unaligned intervals, ambiguous beat anchors, insufficient media/fades, tracked artifacts and previously retimed mapping must reject without silently dropping settings. Depth requires regeneration. Explicit --timeline-settings clear removes old settings with a report; speech-setting migration remains unsupported. Compare and review the exact full MP4; migration is not perceptual approval. This development feature is not in alpha.7.
For development semantic caption grouping, read docs/CAPTION_GROUPS.ja.md. Use session caption-source on a sealed speech render, then session caption-groups --spec-file to propose exact source word-ID/occurrence groups. Interpret requests such as keeping negation, names and units together using actual words; preserve complete ordered coverage and record reasons. Set ja/en, exact protected phrases and editor-selected reading thresholds; never use example IDs or invent timing/spelling. Unknown/missing/reordered refs, stale spelling and groups crossing edit junctions reject. Caption times follow actual cut-word times, retimed word frames or J/L audio samples. Review reading/overlap/punctuation diagnostics, the retained SRT and exact resulting MP4; diagnostics and agent grouping are not human approval. Regular YouTube uses separate SRT; burn captions into vertical MP4 through the existing export and review route when requested. Preserve caption-evidence.json, its SRT SHA and sealed delivery copy. Old renders without word evidence require a fresh render. This feature does not provide full semantic understanding, JLREQ or platform-UI compliance and is not in alpha.7.
In a post-alpha.7 checkout exposing SFX audio_retime, visual session migration preserves effect audio, including loops, through the continuous picture ramp. Original cue fades are applied before pitch-preserving stretch; freezes insert silence and resume source PCM. BGM remains normal speed and ducking is calculated against the new speech clock. Keep source/trim/fade/mapping SHA bindings and content-retime evidence. Changing the original fade or trim requires a fresh original candidate; do not remove these bindings to force reuse. Looped SFX binds its original decoded period and seam taper in version-2 metadata, repeating on the original clock before stretch; do not restart loops on the new clock. Version-1 nonloop metadata remains readable. Composed SFX retime is not supported. Use baked mix/video_only handoff; editable FCP must reject normal-rate substitution. Synthetic sample/pitch tests do not replace full human listening or device playback review.
In a post-alpha.7 checkout with video_loop_periods, video loop cues repeat only the selected source trim after one output-CFR conform. Keep the actual period-frame count, source/period SHA and sealed RGBA layer. Sealed version-2 phase replay preserves loop position and baked video fade/alpha; never replace old loop content with a newly started loop. Visual cue fades apply alpha. In development first-retime migration, image/static title fades select sealed version-2 RGBA samples from the original output clock, preserving alpha during holds and omissions. Keep source/image or generated title raster SHA, text, placement, opacity and fade bindings. Old static renders without samples allow only complete one-to-one translation; otherwise render a fresh original. Changing loop, trim, placement, opacity or a nonzero video fade requires a fresh original layer. Inspect entry, seam, exit and any repeated gesture. Editable FCP must reject unrepresented loops/visual fades; use baked mix/video_only. Technical synthetic checks do not approve aesthetic repetition or real-device playback.
For development effects comparison adjustments, read docs/COMPARISON_ADJUSTMENTS.ja.md. The page exports version-2 choices with existing-event strength/off operations; unchanged choices keep version 1. Import through session select-comparison, render the unadopted candidate and review the resulting exact SHA before adoption. The page itself does not rerender. Undo/reset affects staged changes only. Captured phase/content-bound retime effects are not adjustable here; choose an original candidate and regenerate instead of discarding bindings. Version 2 keeps strength/off only. Development version 3 additionally accepts exact output-frame ranges for untracked effects and normalized smooth-zoom anchors; BGM/target changes and interval-only rendering remain separate. This is not in alpha.7.
For a development version-2/3 comparison correction, render both original and revised candidates fully, then use session preview-effects SESSION REVISED_RENDER --effect-ids ID --context-frames 8 to inspect matched finished-picture excerpts with audio. Omitting IDs includes all changes; removals retain their original intervals. Keep parent full-render SHAs and frame/source evidence, and record feedback/review against the revised full SHA. Reject stale mappings, previews-only inputs and excerpts over 30 seconds; split IDs rather than silently shortening an effect. This exports excerpts after a full render, not an interval-only rerender, and does not constitute whole-video listening or human approval. This is not in alpha.7.
For development version-3 comparison range/position controls, use the page's observed output FPS/frame bounds; never round rational frame times or retarget tracking artifacts. Smooth-zoom anchors are normalized 0..1 and other parameters remain prohibited. Render before adopting and inspect subject/caption collisions. Interval previews include both old and new windows, including the place where the old effect was removed. Read docs/COMPARISON_ADJUSTMENTS.ja.md. This is not in alpha.7.
For development effect-window trials, use session preview-changes SESSION CANDIDATE_ID after importing a version-2/3 comparison correction. It reuses the actual retained pre-effects SHA and finished audio, preserves original effect clocks/history and grade/encode order, and encodes only selected output frames. It still processes the prefix and can differ from full CRF pixels at GOP boundaries. Keep preview.json bindings; partial artifacts never register as full renders or satisfy adoption/review. Render fully before adoption and inspect that exact SHA. Missing/stale input, non-effect changes and windows over 30 seconds must reject. See docs/COMPARISON_ADJUSTMENTS.ja.md; this is not in alpha.7.
For development compound dance edits, use actual session source ranges and measured music beats; explicitly disclose reordered choreography and added music on a comparison baseline. New visual retime settings use version 2, sealing actual shot boundaries separately from timing fragments. Fresh trails may cross a speed fragment within one shot but must reject actual cuts. Preserve version-1 behavior when resuming older candidates; never infer new boundaries from old fragments to force acceptance. Keep BGM at normal speed when only picture changes are requested. Low-FPS ramps can duplicate/omit frames and do not imply optical flow or smooth-motion approval. Protect observed picture/subject regions and inspect zoom, trail and title together. Deliver separate sound-bearing Before/After artifacts with exact source/output SHA and full AV decode; human whole-video watch/listen remains separate. Report TikTok API configuration and real native music/effect execution independently; neither browser login nor local look-alike effects proves native execution. See docs/RETIME.ja.md. This is not in alpha.7.
For development comparison speed requests, untracked smooth zoom and saturation controls offer shorter/original/longer effect time. This changes rise and return together through the existing version-3 frame-range proposal, preserving the current midpoint and peak strength/anchor. Original duration is the reference for each preset; repeated clicks do not compound. Frame parity keeps an exact midpoint, so display actual resulting frames. Out-of-bounds or sub-minimum choices reject without clipping. Undo/reset and Off remain available. This changes effect duration, not footage/BGM playback speed or independent rise/return timing. Phase/content-bound and tracked effects do not expose it. Import, preview/render and review exact resulting SHA; clicking does not rerender or approve. See docs/COMPARISON_ADJUSTMENTS.ja.md; this is not in alpha.7.
For development BGM corrections, the effects comparison page exports version 4 for existing music cue gain/off. It preserves source-audio settings, SFX cues, music selection/trim/fades/duck/timing, picture and mapping; final normalization can change the whole mix's level. Gain controls require explicit no-normalization or sealed nonzero source-audio RMS; normalized music-only/unknown cases expose Off only because final normalization can cancel scalar gain. RMS does not establish speech or listening quality. Undo/reset preserves each candidate's staged changes. Captured cue/effect clocks remain guarded. Import through select-comparison, render fully and inspect the exact new audio; cached effect previews reject music changes because they retain old audio. This is not track replacement, native TikTok processing, instantaneous playback adjustment or alpha.7. See docs/COMPARISON_ADJUSTMENTS.ja.md.
For development track replacement, comparison version 5 selects a current authorized registered music SHA and exact rational source start. Keep cue duration, output placement, fades, loop and duck; insufficient source rejects without automatic looping. Existing captured cue/effect clocks and beat anchors remain protected. Render fully, preserve music_change_evidence and reanalyze the new track before claiming beat sync; retained cut/effect positions are not new-song beat alignment. Gain/off version 4 remains unchanged. This is local asset mixing, not native TikTok/API processing or alpha.7. See docs/COMPARISON_ADJUSTMENTS.ja.md and docs/GENERATED_DEMO_MUSIC.md.
For development music-bound accents, use session beat-effects on a full music render with the exact analyzed asset SHA/byte count and cue ID. Select actual mapped beat indices and odd frame windows; their discrete midpoint is the peak. Rebuild cuts before ramps, then add fresh accents to the resulting mapping. Keep captured music clocks, protected intervals, complete windows and overlap guards. Reduced motion omits new zoom and limits saturation; it does not remove existing ramps/effects. Preserve motion_template evidence through render/delivery. Automatic beat tracking does not establish strong beats, chorus starts, action peaks or listening approval. This is local rendering, not native TikTok processing or alpha.7. Read docs/BEAT_EFFECTS.ja.md.
For actual external native finishing, retain separate downloaded baseline/result MP4s and use native-compare BEFORE AFTER --output DIR. Check full decode, exact native PCM/sample counts and frame timestamps before comparing reduced picture changes. Unequal timestamps must not be silently aligned; reduced RGB equality is not full-resolution equality. Report Studio UI execution separately from Business/Symphony API response evidence. Audible sound does not identify the selected song or establish cross-platform rights. Use the TikTok API guide; keep natural and requested-effect alternatives separate.
For a requested moment such as “accent near 4.35 seconds,” use session beat-effects --beat-map-file ACTUAL_MUSIC_MAP --cue-id EXISTING_MUSIC_CUE --at 4.35 instead of building an effect JSON. The development time controls select a uniquely nearest quantized music peak within the explicit maximum distance, default 150ms. Missing, ambiguous, duplicate or protected selections must reject rather than jump to another beat. Keep the exact selected/requested times and distance in the sealed evidence. Supply observed protected intervals for speech/procedures; an empty list is not automatic speech protection. Preserve cuts, BGM and mapping and render fully before review. This does not detect a chorus, strong beat or action peak. See docs/BEAT_EFFECTS.ja.md.
For development FCP handoff, preserve original mono asset channels while exporting mono/stereo sources into a stereo project output. FCP 12.4 warned on a top-level mono project layout. Use check-fcp --allow-media-relocation only when actual local source files are available; it requires SHA/byte equality for copied assets and keeps rational clip timing checks. Do not substitute filename/uid matching. Native returned-XML mapping and minimal warning-free GUI import are established separately from playback, audio/picture parity and complex editable automation. Consult docs/FCP_ROUNDTRIP.ja.md; the complex calibration import remains unaccepted.
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.

