Open Film Skills

Portable, modular directing intelligence for Codex, Claude Code, TRAE, CodeBuddy, WorkBuddy, and other Agent software.

简体中文 · 日本語 · 한국어 · English

Two ways in

Start here

The design guide is the human-readable entrance. The linked runtime file remains the exact Agent instruction, and the ZIP remains the standalone installable package.

Testing: storyboard motion candidate 5.4.3

ai-storyboard-director-motion-lab is an explicit-only experimental test line for complex camera paths, motion-reference video, visual transformations, and 3D previs responsibilities. It does not replace formal ai-storyboard-director 5.4.2, is excluded from the complete-studio ZIP, and requires --experimental for local installation. Read the runtime contract · Download the experimental ZIP

The promise

These Skills do not replace judgment with prompt decoration. They turn story causality, character purpose, blocking, spatial continuity, physical action, light, materials, and production gates into reusable operating contracts.

The visible result comes first: a readable script, shot plan, asset contract, visual direction, research report, or qualified media deliverable. Internal schemas and checks support that result; they do not replace it.

Read the design before installing

All 20 modules have a GitHub-readable design guide in both English and Simplified Chinese: 40 pages generated from one reviewed contract registry. Every page explains the module's purpose, principles, inputs, workflow, directed return path, review gates, pass evidence, outputs, boundaries, host requirements, and every file shipped in that standalone package.

• Browse all English and Chinese module guides
• Understand the shared return, review, and pass logic
• Compare modules by task, runtime source, ZIP, and evidence state

The guides explain the runtime contract; they do not replace it. SKILL.md remains the canonical Agent instruction. Structural validation, host execution, real-task evidence, and explicit user acceptance remain separate states.

Quick install

Clone the repository:

git clone https://github.com/62656456/ai-film-skills.git
cd ai-film-skills

See every available module:

python scripts/install_skill.py --list

Install one Skill:

python scripts/install_skill.py ai-storyboard-director --platform claude-code

Or download one ready-to-extract ZIP from the latest release. Each archive contains one complete Skill folder.

Choose the host explicitly:

python scripts/install_skill.py ai-storyboard-director --platform codex
python scripts/install_skill.py ai-storyboard-director --platform claude-code
python scripts/install_skill.py ai-storyboard-director --platform codebuddy

TRAE uses a project-level .agents/skills/ folder. WorkBuddy imports the release ZIP through Add Skill → Upload Skill. Every packaged folder carries its own required references; experimental Skills are excluded from the normal complete package. See Installation and Agent compatibility.

Designed to travel alone

• Every package contains a portable SKILL.md plus its own scripts and required references. agents/openai.yaml is optional Codex metadata and is never a runtime dependency.
• Every module is independently useful inside its stated outcome boundary. “Standalone” does not mean that a style module writes a script, an asset contract generates an image without a media tool, or a production workflow bypasses cost and permission gates.
• Each design guide states what the module can deliver alone, what it cannot claim alone, and which host capabilities are still required.
• The repository validator rejects missing local dependencies and a return of the old shared-reference folder.
• Releases publish one ZIP per Skill plus one complete-studio ZIP and a SHA-256 manifest.
• Structural portability is checked automatically; real-project quality remains a separate status claim.

Try a module immediately with the quick-start prompts.

Works with your Agent

The canonical content is shared across every host. Native discovery and tool permissions still belong to the host, so the repository distinguishes native support from a prompt-only fallback. The exact product-name checks—including “Cloud Code,” “Trint,” and “WorkerBilly”—are documented in Agent compatibility.

Skill map

See the detailed Skill catalog, 40 design guides, shared design system, and architecture.

Status means something

• Deployed: the current personal runtime package is in use. It is not automatically described as stable in practice.
• Packaged: structurally complete enough to distribute, but not currently deployed.
• Experimental: isolated from normal installation and clearly marked as unapproved or incomplete.
• Retired: intentionally absent. A retired package is not restored from an old file.
• Third-party: not republished as original work.

Repository design

The repository is organized as a production map rather than a wall of prompts. The visual system uses film-slate black, script-paper white, signal orange, cool cyan, and brass. One horizontal route—story → assets → visual language → shots → production → validation—acts as the signature element across the README and diagrams.

The information architecture was informed by OmniRoute: a strong visual thesis, immediate navigation, visible status, multilingual entry points, copyable quick starts, diagrams, contribution routes, security guidance, and explicit third-party notices. No OmniRoute brand asset, illustration, copy, or code is included here.

Contributing and safety

• Read CONTRIBUTING.md before proposing a Skill change.
• Report accidental secrets or private information through SECURITY.md, not a public issue.
• See THIRD_PARTY_NOTICES.md for excluded and adapted material.
• Check the exact public boundary in PUBLICATION_SCOPE.md.
• Regenerate the bilingual guides with python scripts/generate_skill_guides.py after changing the reviewed contract registry.
• Run python scripts/validate_skill_docs.py and python scripts/validate_repository.py before submitting changes.

Feedback and contact

The project grows through concrete use, not promotional claims. Share the request you tried, the Skill used, what worked, and the one improvement that would matter most.

• Discuss workflows in GitHub Discussions.
• Report reproducible problems or proposals in GitHub Issues.
• Contact the maintainer at [email protected].
• Use the structured feedback guide when possible.

License

Personally authored content in this repository is licensed under the Apache License 2.0, unless a file says otherwise.