Prompt Diff Feed Skill

What you get

Six stable split-diff scenes
Two original prompt and configuration briefs
Measured SVG layout, SRT captions and local audio support

You will need

Node.js 22+, FFmpeg/FFprobe and pinned HyperFrames 0.8.181
A source-backed English brief with exact text pairs
Optional local Kokoro environment; silent output is supported

Good fit

Prompt edits, configuration reviews and instruction handoffs
Small changes where the old and new text must remain visible

Not for

Claims that a wording change improves a model
Untraceable or invented before-and-after text
Cloud media generation or third-party UI captures

You say

“Use Prompt Diff Feed to show a precise prompt or configuration change. Provide before, after and preserved context; describe the wording change only and never infer a model-quality improvement without evidence.”

After installing, say something like this to your agent with your own topic and assets. It asks only for what is missing.

FAQ

Do I need an account?

Yes. Members Skills install from a signed-in FeedSkills account with an active Founding Membership.

What does it cost to run?

The Skill is included with Founding Membership. It needs the tools listed under “You will need”. Any voice, model or media service you choose bills you directly; FeedSkills does not include provider credits.

Can I use the videos commercially?

Videos you make are yours to use, including commercially. The Skill package is licensed as: FeedSkills Source-Available 1.0 · bundled fonts retain SIL OFL 1.1. Fonts, music, footage and services you add keep their own terms.

Which agents does it work with?

Agents that read SKILL.md files, such as Codex or Claude Code. What FeedSkills has actually tested is listed under Tested scope below.

Tested scope

Version: 0.1.0
Acceptance: This version has a recorded acceptance entry.
Tested: Two 20.4-second 1080x1920 / 30 fps English example films and one single-scene edit were rendered with burned captions, measured DOM layouts and full-stream checks; both themes recorded 18 passes, 1 warning and 0 fails. The edit rerendered evidence-edit and reused the other five scenes with native-frame proof; package check/create and DOM guard passed. Real-phone review and customer adoption remain separate.
Runs on: Claude Code · Codex · macOS · Node.js 22+ · FFmpeg
License: FeedSkills Source-Available 1.0 · bundled fonts retain SIL OFL 1.1

Details

Prompt Diff Feed

Prompt Diff Feed is a Members Skill for showing exact prompt or configuration edits while preserving surrounding context. It describes wording changes only and does not claim a model-quality result.

Reviewed A/B previews

Theme A video · cover (absolute video)
Theme B video · cover (absolute video)

Both examples use six stable split-diff scenes with measured layout, burned captions and SRT. They show distinct English edits and include a local single-scene revision; the wording examples are not before-and-after performance tests.

Install and use

Install the approved member package with:

npx --yes --package=https://feedskills.com/downloads/feedskills-0.3.6.tgz feedskills install prompt-diff-feed --agent codex

Copy examples/theme-a.json or examples/theme-b.json into a project outside the installed Skill. Preserve the before text, after text, context lines and source notes exactly, then run project.mjs doctor, project.mjs check, project.mjs create and the generated render entry. Use local Kokoro for narration; node render.mjs --no-voice is an explicit silent export.

For a single-scene revision, keep scene ids and durations stable, edit one copied diff, rerun the render and inspect render-report.json. Compare the changed scene with native decoded frames for the five unchanged scenes.

Boundaries

Do not infer a better model, stronger instruction or safer configuration from a text diff alone. Keep exact whitespace and source notes, avoid credentials and never use remote UI captures or paid media generation.