AchLim/video-prompt-director

### An Agent Skill for Codex and Claude Code that turns scripts and approved storyboards into AI video prompts a model can perform.
Camera direction, timed action, natural acting, exact dialogue, continuity between shots, and a quality check before each prompt reaches you.
Install · What it does · Before and after · Models · FAQ
## What it does
Ask your agent for video prompts and video-prompt-director plans a storyboard and maps its panels to generations before writing prompts.
Storyboard first. Give it a script, a scene, a shot list or an existing storyboard. If there is no approved storyboard it proposes panels and waits for your changes before it writes prompts.
Prompts that direct a performance. Camera and subject movement are described separately. Emotion is written as visible behavior, so people do not freeze, stare or smile through a scene.
Audio and references. Choose native audio, dialogue-only, supplied recordings or silence. Bind subject, motion, camera, style, audio, scene and clay references separately, with version-specific uncertainty.
Editing and assembly. Plan targeted video and audio edits, extensions, transitions and multi-asset workflows, including what to preserve and how to verify it.
Exact dialogue. Your lines stay word for word, attributed to the right speaker, with the listener reacting.
Continuity across shots. Identity, wardrobe, props, positions, light and screen direction carry from shot to shot, with a ledger when you want to check it.
A check before you see anything. Every prompt is validated for action, progression, plausibility, contradictions and fit with the target model, then fixed.
A quantified camera. Lens and depth of field, angle, composition, one slow move chosen for a story reason, and stated stability, so the model does not guess.
Rules out what models add on their own. No generated text, subtitles, logos or music unless you ask, no extra people, no cloned faces, no wardrobe drift.
Model files with sources and dates. Seedance, Veo and Flow, and Runway each have a file that states what is documented, where, and when it was checked. Kling 4.0 and 3.0 are covered the same way. The skill does not invent model limits.
It does not generate video, and it does not change the code of the project you have open.
## Before and after
Weak
A woman stands in a living room looking sad.
With the skill
Medium shot, eye level, locked-off camera. A woman stands beside the living-room window, looking outside. She lowers her gaze and takes a restrained breath. Her right hand tightens around the curtain. After a pause she turns her head toward footsteps behind her, and her expression shifts from sadness to quiet apprehension. Natural breathing, small shoulder movement. She ends facing the doorway.
The first gives a video model a pose. The second gives it a performance with a start, a change and an end. More pairs are in examples/weak-vs-strong-prompts.md.
## Install
Pick one route per host and scope to avoid duplicate copies of the same skill.
### Codex
Copy the complete skill folder to /.agents/skills/video-prompt-director/ for one project or ~/.agents/skills/video-prompt-director/ for all projects. With the npm CLI, use --dir to select the parent skills directory: