菜单
Review a generated image/keyframe for brand fit, accidental in-frame text, claim/restriction violations, subject consistency, aspect, and quality artifacts BEFORE committing budget. Use after a preview/keyframe and before finalizing or generating clips. Keywords - qa, review, check, on-brand, consistency, looks right, before publishing.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | qa-review |
| description | Review a generated image/keyframe for brand fit, accidental in-frame text, claim/restriction violations, subject consistency, aspect, and quality artifacts BEFORE committing budget. Use after a preview/keyframe and before finalizing or generating clips. Keywords - qa, review, check, on-brand, consistency, looks right, before publishing. |
| allowed-tools | Bash Read Write Edit Glob Grep |
Catch off-brand colors, accidental text in frame, restriction violations, character drift, wrong aspect, and artifacts before you pay for the full render or the video clips. Cheapest place to fail is a 512px preview / a $0.067 keyframe — not a $14 video.
Every workflow function is exposed as a CLI command:
node workflows/cli.cjs <command> '<json-args>' # node workflows/cli.cjs list shows all commands
generateImageOptions) before finalizeImagePull the brand facts from brand.md (colors, tone, restrictions.never_say) and the
expected subject from the registry/plan, then review:
# brandColors from brand.md; restrictions from brand.restrictions.never_say; expectedSubject is the locked character
# checks defaults to all: ['brand-colors','unwanted-text','claims','consistency','aspect','quality']
node workflows/cli.cjs reviewOutput '{"imagePath":"projects/{name}/output-contents/storyboard/scene-01-struggle.png","brandColors":["#2563EB","#10B981"],"brandTone":"professional, warm","restrictions":["guaranteed","cure","risk-free"],"expectedAspect":"9:16","expectedSubject":"late-20s East-Asian woman, beige knit, gold hoop earrings"}'
# data: { pass, score, issues:[{check, severity, detail}], summary }
pass === true → proceed (finalize / generate clips / register).
pass === false → there is an ERROR-severity issue. STOP and show the user:
• the issue list (check + detail)
• offer to regenerate with a corrected prompt addressing each error
• re-review the new output before spending
warnings/info → surface them, but the user may accept and proceed.
Always show the user the issues and the score — don't silently pass or block. A failed QA on a cheap preview is the system working: it just saved a wasteful expensive run.
Reviewing a video? Extract a representative frame first (
ffmpeg -y -i clip.mp4 -vf "select=eq(n\,0)" -frames:v 1 frame.png) and review that.
generateImageOptions → reviewOutput(each) → user picks a PASS → finalizeImage → reviewOutput → ship
generateStoryboard → reviewOutput(each keyframe) → fix fails → approve → generate clips
Video QA is one call now — no manual frame extraction:
node workflows/cli.cjs reviewVideoOutput '{"videoPath":"…","expectAudio":true,"expectedDurationSeconds":N,...}'
Samples frames + checks container (duration/audio). For kinetic/typography
content pass checks WITHOUT "unwanted-text"/"claims". Run after EVERY video
generation; fix what frames[] flags before packaging.
MANDATORY pre-flight check that runs BEFORE any content plan or generation. Classifies the topic, verifies project/brand/product context, resolves the required reusable assets (characters, products, logos, voices) from the asset registry, and provides/generates whatever is missing — saving paths for reuse. Use whenever the user asks to plan a campaign, make a marketing plan, or create/generate any image, video, voiceover, or music.
Generate music using Gemini Lyria. Use for background music, jingles, intros, soundtracks, and complete songs WITH sung vocals — Lyria 3 sings your exact lyrics (8 languages, rap/sung delivery).
Generate images using Gemini 3 Image (Nano Banana). Use for thumbnails, carousels, infographics, social graphics, product photos.
Generate video content using Gemini Veo. Use when user asks to create video, TikTok ad, Instagram Reel, YouTube video, product video, explainer.
Generate voiceover audio using Gemini TTS. Use for narration, voiceovers, podcasts, multi-speaker dialogue.
Write the words — scroll-stopping hooks, framework-driven scripts, and platform-ready captions with hashtags. Use when the user needs hook ideas, a video/ad script, ad copy, post captions, or says "what should I post", "write a script", "give me hooks", "write the caption".