Prompt Preflight QA

Follow shared release-shell rules in:

• postplus-shared release-shell rules

Use this skill immediately before image or video generation.

This skill is for:

• checking whether a prompt draft is structurally strong enough to run
• finding missing constraints before credits are spent
• pointing drift risk back to the right upstream stage
• producing concise rerun notes

This skill is not for human creative approval after outputs exist.

Core Rule

Judge prompt quality by controllability, not by how impressive the prose sounds.

Method

Read references/checklist.md.

Default Workflow

1. Classify the prompt object

Choose one:

• storyboard grid prompt
• full video request
• no-reference draft
• reference-led replication draft

2. Run checklist review

Check:

• opening strength
• chosen hook mechanism is legible in the first 3 seconds
• viewer question clarity
• visible evidence
• promise-delivery match by the next few beats
• product timing
• ad-detection risk from premature product exposure
• negative constraints
• reference contract
• explicit reference bindings when references are used
• realism and UGC control
• output-format correctness
• segment plan correctness when the target Seedance request exceeds 15 seconds

3. Print a preflight report

Use this shape:

Verdict:
Major risks:
- ...
Missing fields:
- ...
Likely drift:
- ...
Fix now:
- ...
Can run now:

Blame Rule

When a prompt is weak, point the problem to the right stage:

• bad routing -> pattern-router
• missing or misfit hook mechanism -> hook-design
• weak hook decode -> reference-decode
• weak panel logic -> storyboard-grid-writer
• bad request mapping -> video-request-architect
• bad learn/do-not-copy rules -> reference-contract-builder

Failure Mode

If the prompt could still run but is risky, do not silently pass it. Mark it as risky and say what is most likely to fail first.

If a Seedance request exceeds 15 seconds and no segment plan exists, mark it not_ready.