Workbench Operations
Convert explicit editing requests into existing CLI operations, always reading back results within the same project. This is not a separate editing engine, nor is it a mandatory entry point before using other skills.
First, Identify the Action
| What the user wants | This skill's approach |
|---|---|
| View the current project, find a phrase, see existing clips and pictures | Read-only positioning; see Operations |
| Add an animation or video over a phrase while preserving the original voice-over | Overlay clip; timeline does not extend; see Operations |
| Delete a picture, word, or clip | Distinguish between object and audio impact; see Deletion Methods |
| Split into two segments, shorten a segment, reorder existing clips | EDL modification; see Operations |
| Change aspect ratio (landscape/portrait), fill or fit the frame | Distinguish between project canvas and media fit; see Operations |
| Insert new material at any position on the main timeline and shift everything after | First check current capabilities; if no generic insert command exists, clearly state the limitation—do not misrepresent overlay or move operations as inserts |
When the user says "insert", first determine whether it means overlaying a picture while the voice-over continues, or adding new duration and shifting everything after. When the user says "delete", first determine whether it means removing a picture or cutting out the audio and time. If the intent is clear, execute without repeated questioning. For duplicate phrases, reused material, or non-unique targets, list all positions for confirmation; do not guess based on text alone.
Automatic detection of speech errors, duplicates, or which version to keep should be handed to the Voice-Over Trimming Skill. Subtitle content and layout go to the Subtitle Skill. Shot/creative animation goes to the appropriate Design Skill. Final output goes to the Export Skill. Connection issues and upgrades go to Maintenance & Troubleshooting. Do not automatically launch additional agents or install missing skills.
Standard Operation Flow
- Connect: On first connection or when credentials change, read Connection and Submission in full. Retain the confirmed Runtime parameter array, explicit origin, server projectId, and capabilities—do not rediscover the environment for every action.
- Read: Read only the relevant documents; save the document's own revision, target ID, original values, and expected results. If locating words, paginate and read fully when necessary. Confirm source identity and instance; do not confuse subtitle cue IDs with word IDs.
- Plan: Briefly describe which object will be changed, whether the timeline will be shortened, and whether audio will be affected. Do not repeatedly confirm scope that the user has already authorized. New impacts (e.g., additional removal of dependent layers) must be explicitly authorized first.
- Submit: Follow the actual fields and patterns of
workbench commands --json. Use--confirmedon write operations to indicate that authorization is already granted—this is not a switch that grants permission to the agent. Edit the production project only through the Runtime; do not modify the project JSON directly. - Verify: Read back the affected resources; check IDs, differences, and duration. If necessary, verify word anchors, subtitle status, and picture timing. When performing multiple patches in sequence, only construct the next step after the previous one has been successfully read back. This is not an atomic batch transaction.
- Deliver: Report separately what data was written, whether the picture/audio was actually verified, and any unsupported or unvalidated items. A successful submission does not mean the preview or final output passed. A page not refreshed does not mean nothing was written. Always read back first; do not submit again just to "refresh".
On Failure, Stop Writing
- Conflict: Re-read the current differences. Old approvals are not automatically replayed by replacing with the new revision. Only after confirming the new target and impact, make a new plan.
- Timeout, disconnection, or readback failure: The result may still have been written. Save the original request and identity. Query precisely by the corresponding resource, asset hash, or original operationId. If not found, it doesn't prove the submission did not happen. Do not blindly retry with a different ID.
- Insufficient resources, ambiguous target, missing capability: Report the specific blocker; preserve the existing project. Do not automatically upgrade, restart, re-transcribe, modify registry, or switch to another tool.
- Do not provide fictional commands for generic undo/redo, volume mixing, playback control, or arbitrary timeline insertion of new media.
playbackis for reading the playback word stream;operation-recoveris for limited transaction recovery, not arbitrary undo. - Deletion only affects the specified object within the project; do not clean up original videos, recordings, imported assets, module files, or caches.
Typical Requests
- "Overlay this MP4 on the second sentence; keep the audio as is." → Locate the real word ID, import the media, add a new overlay while keeping the old one, read back.
- "Remove the small black animation I just added." → Locate the unique layer, remove only that layer; do not trim the voice-over, do not delete the animation source file.
- "Delete the third occurrence of this sentence and join the rest." → Verify the third instance, follow the supported word/clip deletion flow, and verify the new duration.
- "Split the first segment in half, then move the second half to the end." → Use clip partial cut followed by second-based splitting, read back the new IDs, then move according to magnetic timeline rules.