Current status
Cut Room Handoff is a Round 1 native UGOS Skill draft, not a deployed NAS application. Version 0.1.0 has passed the official static checker with zero errors or warnings. Native installation, delegation, permissions and NAS saves remain untested.
Inspiration
Narration projects can collect scripts, plans and takes from different people or agents. An existing file or matching filename does not establish which generation records belong together. The next contributor needs the relevant source evidence and a clear task, not just a completion claim.
What it is designed to do
For one to three selected narration segments, the Skill reconciles a script, narration plan, selected-take records and producer-sanitized generation metadata. It distinguishes supported contradictions, confirmed missing records and unresolved evidence, while preserving an explicitly confirmed pronunciation rule.
The output is an evidence-rich handoff containing the relevant text, seeds, source references, blockers and next decisions. A concrete narration-request proposal is prepared only when target text, pronunciation and a new seed are explicitly designated. It is not authorization to generate audio.
How the draft is built
Original Markdown orchestrates registered UGOS experts. file-agent uses supported ug-filesystem capabilities for scoped reads, stat, exact-name search and small-record SHA-256 checks. knowledge-agent uses ug-knowledge to save the previewed handoff only through the native approval workflow, followed by scoped readback.
The draft requires safe metadata views, complete bounded evidence, regular-file checks, ordered digest observations and separate output-verification allowance. Unknown native contracts remain unresolved rather than being guessed. No custom executable runtime, external API integration or credentials are included in the Skill ZIP.
Local validation and lessons
A separate local Python experiment reused existing metadata for 47 beats without manually re-entering asset rows. The revised helper passed 37 unit tests and 23 controlled scenarios. These were local tests, not execution of the native Skill.
Independent review exposed a blind spot: changing both script and plan could leave an old take selected. Existing generation sidecars supplied recorded input text and seeds for additional consistency checks. They still do not prove the audio bytes or human review.
A separately invoked local agent also consumed a fresh evidence-rich packet, computed its input hash and prepared a dependent narration request without the original conversation. That demonstrates same-machine artifact continuation, not cross-node transport or production media generation.
Limits and next validation
The first version expects a known narration-record schema and producer-side sanitized inputs; no exporter is bundled. It does not inspect audio, infer listening approval, repair source files, dispatch remote workers, render or publish. Human time savings and commercial demand are unmeasured.
If selected, NAS testing will cover a consistent-record handoff with approved save/readback, revised copy with an old selected take, and a denied-read edge case. The native validation matrix also covers incomplete results, changed sources and pending or uncertain saves.
Development resources
The project uses the official UGREEN Skill Developer Package 0.7.3 under applicable sponsor terms. The native ZIP contains original Markdown only. The separate local prototype uses Python's standard library under PSF-2 and component notices; it is not bundled.
Built With
- markdown
- ugos
