agentleFS
ttalkkaklabttalkkaklab/social-flow/skills/generate-cut-videos/SKILL.md

generate-cut-videos

Skill1 starsChanged 3 months agoApache-2.0View on GitHub

Writes each generated_video shot's video prompt, then makes its video after the user approves that shot (or every shot); stops before assembly. Use when videos are requested or the episode order reaches them.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If coverage is wrong, return to the earlier authoring/image stages rather than patching the prompt alone.

    line 18
  2. Very specific:

    Prompt: After every picture shot has its images, write each generated_video shot's prompt with shot_prompt.

    line 21
  3. Very specific:

    Generate: Run video_execute within the saved models and existing budgets.

    line 47
  4. Distinctive:

    After generation, inspect whether those particular corrections appear in the actual clip.

    line 59

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Generate Cut Videos

Steps 7–8 of the production order (user directive 2026-10-31). Read video model selection and duration planning.

For dialogue, read dialogue coverage before writing prompts. Confirm the visible speaking face and silent foreground listener in OTS; two speakers in one clip need readable faces or an explicit creative exception. If coverage is wrong, return to the earlier authoring/image stages rather than patching the prompt alone. Preserve MSR two-speaker bindings and the selected voice lane.

Prompt. After every picture shot has its images, write each generated_video shot's prompt with shot_prompt. Give action (what moves from the opening picture), end (the closing picture) and sound in English; the tool builds the rest from the shot's icon attributes — camera move and speed, focus, and on an MSR shot its size, angle and depth — and quotes its on-screen lines. MSR's lines and image bindings are added by video_plan. The tool records what the prompt came from; when the shot's characters, on-screen lines, frame plan, camera, focus or MSR framing change later, the stage tools warn and the review page marks the prompt stale. Rewrite it with shot_prompt before planning that shot. MSR submits the composed opening image as a background reference; it does not pin that image to frame zero. Do not promise exact framing from that input alone. If the actual result repeatedly changes a required locked setup or duplicates a referenced person, explain the limitation and prepare a scoped first-frame-conditioned alternative. Preserve the saved model and resolution unless the user authorizes a change, and keep attempt limits.

Approval. Plan each shot with video_plan (it sends the stored prompt when args name none, and on MSR the start image rides in the background slot). Before video_execute, show the user that shot's prompt, start (and end) image and cost, and ask: approve this shot, approve every shot from now on, or change something. Record a yes with video_approve {planIds:[…]} and an approve-all with video_approve {auto:true}, each with the user's own words. Never call video_approve without their answer; video_execute refuses a plan without one (video_approval_required). A changed prompt, image or model is a new plan and needs a new yes unless auto is on; video_approve refuses a plan that went stale before the yes (video_plan_stale), so plan it again and show the new one. It also writes storyboard/video-approvals.js, so the review page marks each shot approved, awaiting approval, or changed (prompt, start/end image or lines) after its yes. render-astra-shot.js runs only after the same yes.

Generate. Run video_execute within the saved models and existing budgets; use video_status after an interruption. Import usable existing clips instead of regenerating. For a requested correction, reconcile the opening image, camera endpoints, performance, line acting and prompt before replanning. Remove the superseded action from all of them: an empty-handed report must not retain equipment handling, and a distant recognition must not retain a call-and-response. Repeated camera setups keep their position, lens, background scale and screen sides. Name each visible person's appearance, position and speaking turn; reference panels identify one person, not additional cast members. Trace an exit through visible free space in the opening image before choosing a screen edge. A character cannot leave through a wall merely because the prompt says "exit left". Inspect the entire departure interval, including the edge crossing; correct identity does not establish a valid walking path. After generation, inspect whether those particular corrections appear in the actual clip. A successful receipt or an attached player does not establish correct blocking. Preserve failed versions and distinguish generated, connected and visually verified results. Bind a clip with on-screen dialogue through speech_manage attach_video; an MSR clip binds as native, with no line audio (shot-speech.md adapter layer). Before reporting that the storyboard is connected, open the user's review page and verify that every completed shot has a video player whose metadata loads without a media error. File existence and visual.video.clip alone do not prove the page displays it. The bundled template previews that bound source when productionPlan.cuts is absent; an explicit cut list takes priority. Refresh an older review shell from the bundled template instead of inventing edit cuts just to expose a source clip. Look at each clip before assembly. Stop before assembly; review-cut-videos runs only when the user asks.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.