Video craft
Build something meaningful, on Sui. A superpower for AI coding agents (Claude Code, Codex, Cursor, Grok Build) to ship real products on Sui.
npx -y skills add pivyme/suiperpower --skill video-craftAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 12 stars12 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Polish demo or marketing video frames, pacing, captions, and audio. Use when the user wants to polish a video, do an edit pass, or improve video quality.
SKILL.md
5.6 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it
Preamble (run first)
# Suiperpower telemetry. Routes through the CLI so projects.json + Convex stay in sync.
# Silent on failure so the skill never blocks. Opt out: SUIPERPOWER_TELEMETRY=off.
#
# AGENT NOTE: when this skill finishes, run the matching completion command:
# suiperpower track video-craft ship completed
# Or use "failed" / "aborted" if it ended that way.
command -v suiperpower >/dev/null 2>&1 && suiperpower track video-craft ship started >/dev/null 2>&1 &
true
If TEL_PROMPTED is no, before doing real work, ask the user:
Help suiperpower get better. We track which skills get used and how long they take. No code, no file paths, no PII. Change anytime in
~/.suiperpower/config.json.A) Sure, anonymous B) No thanks
Write the answer to ~/.suiperpower/config.json telemetryTier field and create ~/.suiperpower/.telemetry-prompted. Then continue.
What this skill does
Performs a frame-level polish pass on an already-edited video. Walks five categories in order: pacing, captions, color, audio, and accessibility. Surfaces concrete fixes per category, prioritized by viewer impact. Does not produce or edit footage; produces a fix list the user (or an editor) executes.
The skill assumes a draft video exists. For planning a video from scratch, use marketing-video or the script in submit-to-sui-overflow.
When to use it
- After a draft cut exists, before publishing.
- When the user has feedback that the video "feels off" but cannot articulate what.
- Before re-rendering for distribution, since some fixes change the export settings.
When NOT to use it
- Before a draft exists. The skill cannot polish what is not made.
- For full-flow planning. Route to
marketing-videoor the script step insubmit-to-sui-overflow. - For audio mastering only (a deeper task than this skill covers).
If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.
Inputs
- A reachable URL or local path to the draft video.
- The script and shot list from
marketing-video.mdorsubmission-context.mdif present. - The target channels and their specs.
Outputs
A .suiperpower/video-craft.md with the fix list:
## Video craft, <timestamp>
### Inputs
- video: <URL or path>
- duration: <seconds>
- channel target: <twitter | shorts | landing-page | judges>
### Pacing
- finding 1: <timestamp range, observation, fix>
- ...
### Captions
- finding 1: ...
### Color
- finding 1: ...
### Audio
- finding 1: ...
### Accessibility
- finding 1: ...
### Priority
- P0 (must fix before publish): ...
- P1 (should fix): ...
- P2 (nice to have): ...
Workflow
-
Read the video and the script
- Watch the video end to end at full speed.
- Watch again with the script open. Note where the video diverges from the script.
-
Pacing pass
- Identify dead frames (over 0.5 seconds with no visual change).
- Identify rushed cuts (under 0.5 seconds where the viewer cannot read the on-screen text).
- Recommend trims and extensions with timestamps.
-
Captions pass
- If the video is silent-friendly, captions must carry the full message.
- Check for typos, misalignment, off-screen overflow.
- Confirm captions are burned in (or available as a sidecar SRT for platforms that need it).
-
Color pass
- Check exposure, white balance, contrast across cuts. Multi-source footage often has mismatched looks.
- Confirm brand colors if any are present in the video.
- Flag any text that fails contrast on its background.
-
Audio pass
- Check loudness target: -14 LUFS for streaming, -16 LUFS for Twitter. Use a meter, not a feeling.
- Check voiceover clarity. Plosives, sibilance, room tone.
- Check music ducks under voiceover by 6 to 9 dB.
- Confirm SFX are mixed below voiceover and do not stack.
-
Accessibility pass
- Caption coverage 100% of voiceover.
- No flashing content above 3 Hz.
- Color-only signals paired with shape or text.
-
Prioritize fixes
- P0: anything that breaks the message (caption typo, audio drop, demo flow that does not show value).
- P1: quality issues that materially hurt perception (mismatched color, bad pacing).
- P2: polish items that improve but do not block.
-
Writeback
- Append
.suiperpower/video-craft.md.
- Append
Quality gate (anti-slop)
Before reporting done:
- Was the video actually watched twice (once free, once with the script)?
- Was loudness measured with a meter, not approximated?
- Are findings tied to specific timestamps, not vague impressions?
- Is each fix actionable (a target trim, a target color, a target loudness)?
- Did the writeback happen?
If any answer is no, the skill keeps working.
References
On-demand references (load when relevant to the user's question):
references/loudness-targets.md: Per-channel loudness targets and meter recommendations.references/pacing-rules.md: Frame-level pacing rules with examples.
Use in your agent
- Claude Code:
claude "/suiper:video-craft <your message>" - Codex:
codex "/video-craft <your message>" - Grok Build: run
grok, then/video-craft <your message>in the session - Cursor: paste a chat message that includes a phrase like "polish my video", or load
~/.cursor/rules/video-craft.mdcand reference it.
If you activated this and the user actually wants something else, consult skills/SKILL_ROUTER.md and hand off.
Gives 0 of the 12 instructions most video audio skills give in ~1.4k tokens
Counted across 621 of the 795 authors here whose files we hold, read 2026-08-06
- read individual rule files for detailed explanationsin 21 of 621, across 9 files
- Use WAV PCM 16kHz mono audio formatin 13 of 621, across 4 files
- render final videoin 13 of 621, across 6 files
- use this skill when dealing with Remotion codein 11 of 621, across 4 files
- save generated audio to a WAV filein 11 of 621, across 4 files
- handle conversion errors gracefullyin 10 of 621, across 6 files
- add captions to videos alwaysin 10 of 621, across 4 files
- generate music from text descriptions using MusicGenin 9 of 621, across 2 files
- do not skip pipeline layersin 9 of 621, across 3 files
- do not make one tool do everythingin 9 of 621, across 3 files
- never ask the user to paste their full API keyin 9 of 621, across 3 files
- use azure document intelligence for complex pdfsin 9 of 621, across 4 files
Said here and by no other author read
- run telemetry start command first
- prompt user for telemetry consent if required
- watch the video twice
- identify dead frames over 0.5 seconds
- identify rushed cuts under 0.5 seconds
- check captions for typos and alignment
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.