Creative QC
Use this skill to review creative work before handoff, generation, or publishing.
Trigger Cues
- "review this storyboard"
- "sanity-check these prompts"
- "is this ready to generate"
- "spot issues before we publish"
- "quality-check this creative pack"
Best For
- storyboard review
- prompt pack review
- script cleanup
- checking generated assets against plan
- catching mismatch before spending more generation budget
When To Use
- The user wants review, risk detection, or final sanity-checking.
- There is already a concrete artifact to inspect.
- You need to compare plan vs execution, or prompt vs output.
When Not To Use
- The user wants first-pass ideation or prompt creation, not review.
- There is no concrete artifact yet.
- The user already asked for direct rewriting instead of review-first feedback.
Workflow
- Identify the artifact type:
- brief
- storyboard
- prompt pack
- generated images
- generated videos
- Run structural checks:
- language consistency
- count alignment
- one-to-one mapping between scenes and prompts
- hook strength and opening variety
- CTA presence when the content is promotional
- platform fit such as
9:16, 16:9, short-form pacing, or still-image use
- Run creative checks:
- visual coherence
- tone consistency
- prompt specificity
- repeated or generic phrasing
- If real assets are provided, compare them against the plan:
ImageUnderstandingAgent for images
VideoUnderstandingExpert for videos
- Return prioritized issues and concrete fixes.
- Only rewrite automatically if the user asks for direct revision. Otherwise review first.
Severity Rules
pass: no material blockers, only optional polish
revise: usable direction but there are issues that should be fixed before generation or handoff
block: high-risk mismatch, missing inputs, or structural failure that should stop the next step
Handoff Examples
If a prompt pack needs source comparison first:
{"input_path":"inbox/reference.png","mode":"prompt"}
If a generated video needs intent comparison:
{"input_path":"inbox/output.mp4","mode":"shot_breakdown"}
If the QC result says the script itself should be revised:
{"input_text":"<script to revise>","mode":"rewrite","constraints":"fix repetition and keep the original meaning"}
Output Contract
Return a QC summary like this:
{
"status": "pass",
"critical_issues": [],
"major_issues": [],
"minor_issues": [],
"fix_plan": [
"..."
],
"residual_risks": [
"..."
]
}
Guardrails
- Distinguish objective mismatch from subjective taste.
- Review first; do not silently rewrite the user's material unless asked.
- If there are no issues, say what you checked so the pass result is meaningful.
- When reviewing prompts, keep model compatibility in mind: concise, unambiguous, and medium-appropriate prompts usually beat overloaded prompts.
- Prioritize findings that would waste generation budget, break platform fit, or create scene-to-prompt mismatch.
1---2name: creative-qc3description: Use when the user wants to review or sanity-check a storyboard, prompt pack, script, or generated asset set for consistency, platform fit, and creative quality before final delivery or generation.4---56# Creative QC78Use this skill to review creative work before handoff, generation, or publishing.910## Trigger Cues1112- "review this storyboard"13- "sanity-check these prompts"14- "is this ready to generate"15- "spot issues before we publish"16- "quality-check this creative pack"1718## Best For1920- storyboard review21- prompt pack review22- script cleanup23- checking generated assets against plan24- catching mismatch before spending more generation budget2526## When To Use2728- The user wants review, risk detection, or final sanity-checking.29- There is already a concrete artifact to inspect.30- You need to compare plan vs execution, or prompt vs output.3132## When Not To Use3334- The user wants first-pass ideation or prompt creation, not review.35- There is no concrete artifact yet.36- The user already asked for direct rewriting instead of review-first feedback.3738## Workflow39401. Identify the artifact type:41 - brief42 - storyboard43 - prompt pack44 - generated images45 - generated videos462. Run structural checks:47 - language consistency48 - count alignment49 - one-to-one mapping between scenes and prompts50 - hook strength and opening variety51 - CTA presence when the content is promotional52 - platform fit such as `9:16`, `16:9`, short-form pacing, or still-image use533. Run creative checks:54 - visual coherence55 - tone consistency56 - prompt specificity57 - repeated or generic phrasing584. If real assets are provided, compare them against the plan:59 - `ImageUnderstandingAgent` for images60 - `VideoUnderstandingExpert` for videos615. Return prioritized issues and concrete fixes.626. Only rewrite automatically if the user asks for direct revision. Otherwise review first.6364## Severity Rules6566- `pass`: no material blockers, only optional polish67- `revise`: usable direction but there are issues that should be fixed before generation or handoff68- `block`: high-risk mismatch, missing inputs, or structural failure that should stop the next step6970## Handoff Examples7172If a prompt pack needs source comparison first:7374```json75{"input_path":"inbox/reference.png","mode":"prompt"}76```7778If a generated video needs intent comparison:7980```json81{"input_path":"inbox/output.mp4","mode":"shot_breakdown"}82```8384If the QC result says the script itself should be revised:8586```json87{"input_text":"<script to revise>","mode":"rewrite","constraints":"fix repetition and keep the original meaning"}88```8990## Output Contract9192Return a QC summary like this:9394```json95{96 "status": "pass",97 "critical_issues": [],98 "major_issues": [],99 "minor_issues": [],100 "fix_plan": [101 "..."102 ],103 "residual_risks": [104 "..."105 ]106}107```108109## Guardrails110111- Distinguish objective mismatch from subjective taste.112- Review first; do not silently rewrite the user's material unless asked.113- If there are no issues, say what you checked so the pass result is meaningful.114- When reviewing prompts, keep model compatibility in mind: concise, unambiguous, and medium-appropriate prompts usually beat overloaded prompts.115- Prioritize findings that would waste generation budget, break platform fit, or create scene-to-prompt mismatch.