wyt
65286ee034
Upgrade the raw-render smoke case into a nine-second, three-scene production driven through the built OVS CLI. Prepare/reconcile, snapshot, draft, burn captions and verify the delivered file; assert full decoding, scene order, visible motion, audio across all scenes, actual caption pixels and wrong-duration rejection. Preserve optional review artifacts outside the checkout. Wire this case into pnpm verify, already used by CI. It uses local original graphics and instrumental audio without paid services. Video duration is checked independently from AAC tail padding, within the existing delivery tolerance. Validation: the production case and pnpm verify passed before this commit; all six source sync verification commands passed on the same candidate, with executed-case evidence and content-hash stability. Prompt audit: - Keep the pre-commit/PR production requirement in AGENTS as a resident repository workflow rule: a contributor must know it before invoking Git, where an on-demand production skill may never be loaded. The executable requirement belongs to pnpm verify and the sync checker; AGENTS only defines when to run it and inspect evidence. - Keep command details and retained-artifact usage in the on-demand README; no duplicate production approval flow or Agent/Skill instruction is introduced. The user explicitly requested actual test-video production before submission.
2.7 KiB
Working with OrkasVideoStudio (for coding agents)
This repo lets you produce video by composing, generating, and editing through the ovs CLI
(and an equivalent MCP server). The knowledge for how to do it well lives in skills.
Skills
Reusable, host-neutral SKILL.md knowledge is in packages/skills/. Native skill loaders
discover these automatically:
- Claude Code: install to
.claude/skills/(repo) or~/.claude/skills/(user). - Codex: install to
.agents/skills/(repo) or~/.agents/skills/(user).
If your agent has no native skill loader, pull a skill on demand:
ovs skills # list available skills
ovs skill video-router # print one skill's full instructions
Start every video task by reading video-router; it routes you to the right line
(compose / generate / edit) and the stage skills. Read gate-control before the first
approval boundary so every line uses the same host-neutral transition policy.
Capabilities
The CLI is the canonical interface; the MCP tools mirror it 1:1. Run ovs --help for the
full surface (render / edit / transcribe / narration fit / gate transition / speak /
speech-capabilities / image / video / plan). Run ovs doctor
to check that ffmpeg, ffprobe, and node are available.
Versioning and synchronization
Use one independent calendar version YYYY.M.D across root and all workspace
package manifests; package.json is the source of truth. Use the Asia/Shanghai
release-candidate date without zero padding (for example 2026.9.10), as in
OrkasOpen. Keep the MCP-reported version consistent. Do not copy the Orkas app
version or its marketplace Agent version; dependency and schema versions are
independent. Ordinary changes preserve a valid version until preparing a release.
Record release changes in CHANGELOG.md; release tags use v<version> and an
already published date version must never be overwritten.
Sync from Orkas by reviewed semantic adaptation on a dedicated branch and a PR
to main. Follow the source repository's OpenSource/SyncVideoStudio rules.
Before any Git commit or PR, run pnpm verify on the final candidate. Its real
production E2E must create and decode a three-scene MP4 through the built OVS
CLI, including animation, audio, burned captions and final-delivery checks.
Preserve a sample video and inspect its opening, every scene and ending for
sync/release review; missing runtimes, skipped production cases or failed
checks block submission. A build or container probe alone is insufficient.
After edits, rerun the owning cases and full verification before committing.
Create even a draft PR only
after the requester confirms that concrete result. This does not authorize
merging, tagging, npm publication, or a release.