| name | video-prompt-lab |
| description | Runs iterative prompt experiments for the video creation workflow. Use when Codex needs to test prompts, compare outputs, refine the 8-layer video workflow, update video skills, capture failure modes, or turn a successful from-zero-to-one production into a reusable fixed process. |
Video Prompt Lab
Use this skill to improve the workflow itself. The goal is not one perfect prompt; the goal is a repeatable loop that turns trial runs into stronger prompts, sharper skill boundaries, and a stable production system.
Loop
- Choose a test prompt from
prompts/.
- Run the prompt against the current project state.
- Capture the outputs, mistakes, friction, missing artifacts, and unclear skill routing.
- Update workflow docs, prompts, or the relevant skill.
- Run the smallest useful test again.
- Promote the prompt only after it produces a complete artifact with fewer interventions.
What To Measure
- Did the router pick the right skill and layer?
- Did each layer create the expected file?
- Did any skill do work outside its responsibility?
- Were missing assets, rights, music, SFX, captions, or Remotion timing handled explicitly?
- Did the Remotion structure use Composition, Scene, Sequence, editPlan, and CaptionLayer cleanly?
- Did QA catch real preview/render/package issues?
Artifacts
Use these files:
prompts/master-codex-video-production.md: main from-zero prompt.
prompts/layer-rerun-prompts.md: targeted prompts for one layer at a time.
prompts/retrospective-prompt.md: after-action review prompt.
workflow/prompt-iteration-log.md: running record of prompt versions, tests, and changes.
Revision Rules
- Improve one bottleneck at a time.
- Prefer tightening one skill's responsibility over making the router smarter about everything.
- Add examples only when the failure is repeated or predictable.
- Move detailed reusable guidance into a skill; keep project workflow docs as the operating map.
- Mark the first successful full production as the baseline process before optimizing for speed.
Done Definition
A prompt iteration is done when the test result, observed failure/success, and resulting change are recorded in workflow/prompt-iteration-log.md.