Agent skill

Cut Director

by Fangx-AI in Fangx-AI/cut-director

Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut.

CC-BY-SA-4.0Auto-check passedMedia & Creative

Install Cut Director

skills CLI
$ npx skills add Fangx-AI/cut-director --skill cut-director -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install Fangx-AI/cut-director cut-director --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
cut-director
GitHub stars
179
Token cost
~2.9k tokens
SKILL.md length
1,362 words
Files
265 (incl. scripts, references, assets)
Skills in repo
1
Repo updated
First seen
Licence
CC-BY-SA-4.0

At a glance

Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut.

  • Works in 6 steps: Inspect the source, transcript, timing,… → Load only references needed for this… → Match a verified effect recipe when its… → …
  • A user describes a desired visual effect
  • SKILL.md covers Scope, Core Workflow, Confirmation Boundary and Recipe Routing, plus 5 more sections
  • Asks to make speech-led footage more polished

What it does

Cut Director is an agent skill from Fangx-AI/cut-director. Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut. Use when a user describes a desired visual effect or asks to make speech-led footage more polished, dynamic, cinematic, clear, or visually engaging through visual-beat planning, motion graphics, supporting visuals, speaker composition, reusable CutDirector Prompts, or verified ChatCut execution.

Its SKILL.md is about 2.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 268 other files, including scripts, reference files and assets (for example `.github/workflows/validate.yml`, `CHANGELOG.md` and `CONTRIBUTING.md`).

It sits in Media & Creative, covering Motion graphics and Video production. The repository describes itself as: CutDirector: a ChatCut skill for talking-head editing, with reusable motion graphics prompts and real visual demos. The licence is CC-BY-SA-4.0.

When your agent uses it

  • A user describes a desired visual effect
  • Asks to make speech-led footage more polished
  • Visually engaging through visual-beat planning
  • Motion graphics

Example prompts

  • “/cut-director”

Workflow steps

6 steps, taken from the first numbered list in SKILL.md.

  1. Inspect the source, transcript, timing, frame, speaker, gestures, captions, Logo, product UI, existing text, motion paths, and real empty…
  2. Load only references needed for this request. Identify verbatim anchors and select visual Beats that help the viewer.
  3. Match a verified effect recipe when its viewing task and constraints fit. Otherwise design a custom Beat using the same safety and…
  4. Deliver a confirmable Visual Beat Map and select exactly one representative Beat.
  5. After the first approval, initialize or resume the project manifest and pass every recipe gate before executing only the representative…
  6. Record actual post-write evidence, reach verified, and show the result. Expand only after the second explicit approval.

What it can do on your machine

Read from SKILL.md and the folder at commit d9d2673. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 1 file in scripts/, which the agent can run.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Cut Director loads about 2.9k tokens when it runs, and up to ~64k if it reads all its reference files. Until then it costs about 110 tokens; SKILL.md has 1,362 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~110
When it runs · the whole SKILL.md, loaded when a task matches
~2.9k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~64k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from Fangx-AI/cut-director at commit d9d2673, republished under its CC-BY-SA-4.0 licence (© Fangx-AI). 1,362 words, ~2,938 tokens.

Download SKILL.mdSave it as .claude/skills/cut-director/SKILL.md (or your agent's skills folder). This skill also uses 264 other files; get the full folder from GitHub.
name
cut-director
description
Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut. Use when a user describes a desired visual effect or asks to make speech-led footage more polished, dynamic, cinematic, clear, or visually engaging through visual-beat planning, motion graphics, supporting visuals, speaker composition, reusable CutDirector Prompts, or verified ChatCut execution.

CutDirector

Scope

Direct visuals for already-shot, presenter-led videos. Preserve finished A-roll and original wording by default. Do not use this Skill for a general montage, an unshot script, or a non-presenter-led video.

Let the user describe the result in natural language. Never ask the user to fill an internal schema, recipe, crop parameter, animation curve, or verification checklist.

Core Workflow

  1. Inspect the source, transcript, timing, frame, speaker, gestures, captions, Logo, product UI, existing text, motion paths, and real empty space.
  2. Load only references needed for this request. Identify verbatim anchors and select visual Beats that help the viewer.
  3. Match a verified effect recipe when its viewing task and constraints fit. Otherwise design a custom Beat using the same safety and fallback principles.
  4. Deliver a confirmable Visual Beat Map and select exactly one representative Beat.
  5. After the first approval, initialize or resume the project manifest and pass every recipe gate before executing only the representative Beat.
  6. Record actual post-write evidence, reach verified, and show the result. Expand only after the second explicit approval.

The model owns semantics, director judgment, visual language, and medium choice. Deterministic scripts own required fields, IDs, time ranges, approval state, asset verification state, fallback chains, and evidence completeness. Read references/pipeline-contract.md before execution.

Confirmation Boundary

Do not generate media, create MG, modify the timeline, consume credits, or claim execution before the first approval.

Use existing source context and approvals before asking for anything. Planning, Prompt selection, and read-only inspection do not require execution approval. For a simple requested effect, give a short proposal proportional to the task; do not force a full-video table or reject effect density the user did not request.

If execution approval is missing, show the concrete representative proposal and ask only for the missing decision. If a source or verbatim anchor is missing, ask one focused source question. Never invent timing or facts.

Approvals persist within their actual scope. Record existing approval evidence in the manifest rather than repeatedly asking the same question. A local revision to an accepted Beat can reuse valid direction and scope; update affected facts and verification evidence. New direction, material scope changes, or additional paid actions need the applicable authorization. Do not fabricate the first or second evidence fields or bypass the existing pipeline gates.

Recipe Routing

Load only the matching recipe and its public reference:

User intentInternal recipePublic compatibility path
Official Logo follows a confirmed pointing gesturerecipes/prompt-001-gesture-logo-pop.jsonreferences/prompt-001-gesture-logo-pop.md
Left-side points plus a continuously scrolling long-text evidence columnrecipes/prompt-002-split-screen-explainer.jsonreferences/prompt-002-split-screen-explainer.md
One official brand icon connects two product modes, with progressive capabilities and a final result comparisonrecipes/prompt-003-brand-mode-comparison.jsonreferences/prompt-003-brand-mode-comparison.md
An adaptive progress overlay selects full section tabs, current-section mode, progress-only, or keep-clean from the actual structure, aspect ratio, and safe zonesrecipes/prompt-004-top-chapter-progress-rail.jsonreferences/prompt-004-top-chapter-progress-rail.md
Reuse a verified website-provided page-waterfall-wall.mp4 unchanged; recreate from real screenshots only when no source exists and the user explicitly approvesrecipes/prompt-005-diagonal-card-waterfall.jsonreferences/prompt-005-diagonal-card-waterfall.md
Three original-style light-background cards flip from independently editable front faces to independently editable back faces, with the face swap only at the 90-degree edgerecipes/prompt-006-editable-three-card-flip.jsonreferences/prompt-006-editable-three-card-flip.md
A verified high-resolution real page receives editable typing annotation, a dimming mask, and an accurately positioned focus lockrecipes/prompt-007-hd-page-focus-lock.jsonreferences/prompt-007-hd-page-focus-lock.md
Three to five verified real images fly in as a deck, settle into a fan, and elevate a user-selected hero cardrecipes/prompt-008-real-image-deck-hero.jsonreferences/prompt-008-real-image-deck-hero.md
Editable input, feedback, and result beats explain one causal chain; optional real result media is verified and abstract UI never masquerades as a product interfacerecipes/prompt-009-input-feedback-result.jsonreferences/prompt-009-input-feedback-result.md

Treat recipe triggers as routing evidence, not keyword-only commands. A visual resemblance is insufficient when the viewing task differs.

Apply the recipe's required inputs, asset strategy, safe zones, timing, fallback chain, and verification rules internally. Keep the published Prompt text and paths stable. If a recipe blocks execution, follow its named fallback rather than improvising around the guardrail.

For gesture effects, require a user-confirmed exact time range before asset acquisition or timeline work. If several gestures are plausible, inspect candidate frames and ask only for the exact target range. Do not equate any moving hand with an intentional trigger.

For real brands, use verifiable official assets and never generate, redraw, or stylistically imitate a real Logo. If identity or provenance cannot be verified, stop that asset and request one verified source.

Viewer-task routing

For reusable material, consult the Prompt index. In addition to the existing verified recipes:

  • Music and visual emphasis do not land together: 012.
  • Clauses should accumulate toward one conclusion: 013.
  • Show a real change to the same subject: 014.
  • Connect real demonstration clips within existing speech: 015.
  • Adapt 006/008 to another ratio or add a following focus to 007: variants.

012–015 are local demonstrations, not verified ChatCut recipes. Use as custom Beat references; do not imply native property or real-footage verification. Preserve finished A-roll unless editing it is explicitly within the user’s request. Read compatibility before choosing the current execution surface.

Planning References

Show full SKILL.md (546 more words)Show less
Supplementary Prompt material

For a release/count/viewer-benefit title sequence, consult Prompt 010. For multiple persistent task cards moving through workflow states, consult Prompt 011.

These are local Remotion examples, not verified ChatCut recipes. Use them as custom-Beat references within the existing scope and approval workflow; do not claim ChatCut execution or property editability from the local preview. Their source files and verification limits are linked in each reference. Do not substitute Prompt 010 for a full-video progress rail (004), or Prompt 011 for a single input–feedback–result explanation (009).

For full-video planning, consult these references as needed:

  1. references/visual-director-framework.md
  2. references/transcript-to-beats.md
  3. references/visual-language.md
  4. references/visual-beat-map.md
  5. references/quality-gate.md

Then load only what the selected Beats need:

NeedLoad
Official Prompt lookup or reusereferences/chatcut-official-catalog.md, references/chatcut-prompt-routing.md, references/chatcut-official-prompt-patterns.md
Full-screen, PiP, split-screen, or speaker placementreferences/composition-and-speaker-presence.md
Keywords, lists, charts, chapter cards, or other MGreferences/mg-animation-director.md
Generated visuals, images, or B-rollreferences/generated-visuals-director.md
Full examplesreferences/examples-zh.md or references/examples-en.md

Treat official ChatCut patterns as information-structure and motion references, never mandatory aesthetics, fabricated product UI, or automatic speaker placement.

Director Rules

  • Preserve transcript anchors verbatim. Use approximate or anchor-only timing when exact timestamps are unavailable.
  • Default to 3-8 Beats per 30-60 seconds; fewer or zero is valid.
  • Keep one purpose and one visual focus per Beat. A generated Beat defaults to one continuous shot and one primary camera move.
  • Protect face, captions, gestures, product, Logo, existing text, and motion paths. Lower-right PiP is never a default.
  • Let real product UI become the primary focus whenever visible; keep overlays secondary and non-obstructive. Never fabricate product UI as evidence.
  • Give every reference asset one explicit responsibility.
  • Apply the Quality Gate. Delete or downgrade weak, obstructive, misleading, visually cheap, or unverifiable candidates.

User-Facing Output

For full-video planning, use references/visual-beat-map.md. For one effect or a local revision, present only the affected content, placement, timing, input needs and preview decision. Keep the complete internal evidence without making the user read every field. A full plan includes:

  • overall director judgment and one named visual language;
  • the Visual Beat Map with exact displayed content, speaker treatment, safe zones, editable properties, media/person window, asset responsibilities, compositing, sound, user prompt, director constraints, risks, scores, and quality decision;
  • exactly one representative Beat;
  • segments that should remain clean;
  • high-risk or credit-consuming confirmations;
  • a first-approval checklist covering visual language, speaker treatment, and every credit-consuming action; and
  • the post-approval execution order.

Reply in the user's language. Present the result, not the internal recipe or JSON contract.

Execution And Validation

After the first approval, read references/chatcut-execution-handoff.md and route only the representative Beat to the required ChatCut execution Skills.

Use the internal cache and state flow in references/pipeline-contract.md. Before every ChatCut write, merge known facts and require an executing transition. After the write, record actual asset, beginning, middle, and ending evidence and require a verified transition. For any Beat that covers or replaces the speaker frame, beginning and ending evidence must include the clean frame outside the Beat, the transition in progress, and the settled state; a good middle frame does not prove a clean handoff.

Never expose the manifest, commands, gates, or recovery mechanics as user work. Do not override a validation failure: fix a known fact, apply a documented fallback, or ask for the single blocking input. Show the verified result and wait for the second approval before expansion.

© Fangx-AI, CC-BY-SA-4.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 264 other files (scripts, references, assets) in the repository root of Fangx-AI/cut-director.

  • SKILL.md
  • .github/workflows/validate.yml
  • .gitignore
  • CHANGELOG.md
  • CONTRIBUTING.md
  • LICENSE
  • LICENSES/AGPL-3.0-or-later.txt
  • LICENSES/CC-BY-SA-4.0.txt
  • NOTICE
  • PROMPT-LIBRARY.md
  • PROMPTS.md
  • README.md
  • THIRD_PARTY_NOTICES.md
  • TRADEMARKS.md
  • VISUAL-GALLERY.md
  • agents/openai.yaml
  • assets
  • … and 248 more

Open the folder on GitHubat commit d9d2673

Compare with similar skills

Cut Director next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Cut Director compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Cut Director this skillFangx-AI/cut-director179—~2.9kAutomated safety check: PassCC-BY-SA-4.0
HyperFrames Animationheygen-com/hyperframes60k3 repos~2.1kAutomated safety check: PassApache-2.0
Stitch to Remotion Walkthrough Videosgoogle-labs-code/stitch-skills8.5k6 repos~3.2kAutomated safety check: NotesApache-2.0
Faceless Explainer Videoheygen-com/hyperframes60k3 repos~7.7kAutomated safety check: NotesApache-2.0
Video ShotcraftVincentwei1021/video-shotcraft11k—~2.6kAutomated safety check: PassApache-2.0
Figma to HyperFramesheygen-com/hyperframes60k3 repos~4.5kAutomated safety check: NotesApache-2.0

Similar skills

  • HyperFrames Animation

    heygen-com/hyperframes

    Collects motion rules, scene blueprints, transitions and runtime adapters for HyperFrames video compositions, with GSAP as the default animation runtime.

    60k GitHub starsUsed in 3 repos~2.1k tokens
    Media & CreativeAuto-check passed
  • Stitch to Remotion Walkthrough Videos

    google-labs-code/stitch-skills

    Official

    Builds walkthrough videos from Stitch design projects using Remotion, with transitions, zoom effects and text overlays on each screen.

    8.5k GitHub starsUsed in 6 repos~3.2k tokens
    Media & CreativeAuto-check: notes
  • Faceless Explainer Video

    heygen-com/hyperframes

    Turns an article, notes or a topic brief into an explainer video whose visuals are invented per scene, built frame by frame in HyperFrames with no footage.

    60k GitHub starsUsed in 3 repos~7.7k tokens
    Media & CreativeAuto-check: notes
  • Video Shotcraft

    Vincentwei1021/video-shotcraft

    Makes cinematic product videos with Remotion from shot recipe cards, a ready template, real page screenshots, camera moves and sound design, or builds a single animated shot.

    11k GitHub stars~2.6k tokensUpdated 6 days ago
    Media & CreativeAuto-check passed
  • Figma to HyperFrames

    heygen-com/hyperframes

    Imports Figma assets, brand tokens, components and motion into a HyperFrames video composition, using the Figma REST API with a connector or native export for shaders.

    60k GitHub starsUsed in 3 repos~4.5k tokens
    Media & CreativeAuto-check: notes
  • HyperFrames Video Entry Point

    heygen-com/hyperframes

    Entry point for making, editing and rendering videos from HTML compositions with HyperFrames, routing each request to the right workflow.

    60k GitHub starsUsed in 3 repos~5.2k tokens
    Media & CreativeAuto-check passed

Questions about Cut Director

What does Cut Director do?

Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut. Cut Director is an agent skill from Fangx-AI/cut-director. Direct already-shot talking-head, interview, tutorial, lecture, podcast, knowledge, and presenter-led product videos in ChatCut.

When should I use Cut Director?

Cut Director fits situations like: A user describes a desired visual effect; asks to make speech-led footage more polished; visually engaging through visual-beat planning; motion graphics.

How do I install Cut Director in Claude Code?

Run `npx skills add Fangx-AI/cut-director --skill cut-director -a claude-code`. Or copy the skill folder (the Fangx-AI/cut-director repository) into .claude/skills/cut-director in your project. Claude Code loads it when a task matches its description.

How do I install Cut Director in Codex?

Run `npx skills add Fangx-AI/cut-director --skill cut-director -a codex`. Or copy the skill folder (the Fangx-AI/cut-director repository) into .agents/skills/cut-director in your project. Codex loads it when a task matches its description.

Can I use Cut Director in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add Fangx-AI/cut-director --skill cut-director -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/cut-director, .gemini/skills/cut-director, .github/skills/cut-director and .opencode/skills/cut-director in your project.

What does Cut Director need to run?

SKILL.md names no scripts, command-line tools or credentials: Cut Director is instructions for the agent only.

Does Cut Director access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Cut Director safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Cut Director use?

Cut Director is published under the CC-BY-SA-4.0 licence (from the LICENSE file in the skill folder). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Cut Director use?

About 2.9k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 61k tokens, read only when the agent opens those files.

What are the alternatives to Cut Director?

Skills that share tags, products or a category with Cut Director: HyperFrames Animation (heygen-com/hyperframes, 60k stars), Stitch to Remotion Walkthrough Videos (google-labs-code/stitch-skills, 8.5k stars), Faceless Explainer Video (heygen-com/hyperframes, 60k stars) and Video Shotcraft (Vincentwei1021/video-shotcraft, 11k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Cut Director?

Fangx-AI (a GitHub user) maintains it in Fangx-AI/cut-director, which has 179 GitHub stars. The repository was last updated on October 4, 2026.

Source: Fangx-AI/cut-director on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.