feat(vn): let a finished project hand its media to VN for fine-grained editing

[jiabinchen]

Summary

PAI-Pro generates the raw material for a film — script, characters, locations, voices, storyboards, clips — and then stops at the canvas. This PR adds the missing last mile: an editing handoff to VN, so the generated media can be imported into a VN draft and finished with real editing tools.

The intent of the integration is simple: generate in PAI-Pro, then import into VN for fine-grained editing. It runs fully local — no CawCut account, no upload.

About VN

VN (download) is a cross-platform video editor for macOS, Windows, iOS, and Android. Its editor covers the work a generation canvas does not:

• multi-track editing, picture-in-picture
• text and subtitles, auto captions
• music, effects, AI transitions
• keyframes, speed curves, reverse, frame interpolation
• voice change, image editing, and AI tools

CawCut publishes the cawcut CLI that writes VN's draft format directly. This PR vendors CawCut's cawcut-vn agent skill and wires it into PAI-Pro, so the agent can build a validated VN draft from a project's landed media and hand it to the app for that deeper edit.

What's included

• Vendored skill — skills/cawcut-vn/ (SKILL.md + references/), taken from cawcut/skill v0.6.0. Registered in .claude-plugin/marketplace.json and the PROJECT_AGENT.md skills-routing table, so both Claude Code and Codex projects get it.
• Handoff — story-to-video-workflow now closes a completed run with one choice: open the cut in VN, or inspect the Timeline first. Generation stays single-entry — VN hands media requests back through story-to-video-workflow, never straight to a capability skill.
• Availability detection — before offering VN the agent runs cawcut vn status. "Open in VN" is offered only when VN is installed and meets the minimum version. On an unsupported platform (Linux — i.e. the Docker container) the agent says so and gives the host handoff instead of promising an open that cannot happen.
• Docker — the image installs the cawcut CLI (npm install -g @ubnt/cawcut --ignore-scripts, so its postinstall can't shadow the bundled skill) and zip, so the container can build and pack a draft. It cannot open VN.
• Docs — new docs/vn-editing.md (flow + the Docker docker cp handoff), linked from README, docs/skills.md, docs/setup.md, and the FAQ.

pai-pro-specific adaptations to the vendored skill

The skill is CawCut's; four small edits make it fit here (called out in a pai-pro adaptation section inside the SKILL.md, so a future re-sync from upstream is aware of them):

1. Generation handoffs name story-to-video-workflow instead of the CawCut marketplace skills.
2. Drafts are created inside the project (--dir), not the CLI's ~/Downloads default.
3. Opening is gated on cawcut vn status / platform support (see phase 9).
4. The question tool is mapped for both runtimes (AskUserQuestion → Codex request_user_input), noted in PROJECT_AGENT.md.

Verification

• docker build --check . — no warnings.
• server suite: 366 passing, 1 failing — the failure (require_background_hook.test.js) is pre-existing on main (reproduced there) and is a Node MODULE_TYPELESS_PACKAGE_JSON warning from the workspace root, unrelated to this change.
• Internal doc links resolve.

Notes for reviewers

• Per CONTRIBUTING's "new Skills" carve-out, flagging that this skill is authored and owned by CawCut (MIT, vendored with a version pin), not a community rewrite. Happy to adjust how it's vendored or noted.
• Docker builds and packs a draft but cannot open it — VN is a macOS/Windows app. The host-side open is documented in docs/vn-editing.md (docker cp the .vn out, open it where VN is installed).
• CLA will be signed.

[jiabinchen]

Hi @jeffliangf @Ruijian-Zha-Utopai,

I've just opened this PR to bridge the gap between PAI-Pro and VN Editor.

Why this matters: A significant number of VN users in our community have expressed a strong desire to combine PAI-Pro's powerful generative capabilities with VN's fine-grained timeline editing. This integration delivers exactly that—enabling a seamless "Generate in PAI-Pro, Finish in VN" workflow entirely locally.

We would love to know your thoughts on this integration. Also, please let us know:

1. Your initial impression or intent regarding aligning with VN.
2. Any specific requirements, architectural preferences, or testing scenarios we need to cooperate on during the review/contribution process.

Looking forward to your feedback! 🚀