Two treatments for source-traceable explanations
knowledge-explainer makes a question answerable without hiding the evidence boundary. Choose one treatment for the subject:
- Evidence comparison reads two records as visible documents. Use it for dates, short quotation transcriptions, before/after evidence and a stated difference. A transcription is labelled as a transcription; the document cards and highlights are original schematics, not archival scans.
- Object teardown separates an original Canvas/SVG object into parts and connects each part to its role. Use it for a mechanism such as computing, conduction or energy storage. The drawing is schematic and does not claim exact scale or a complete inventory.
These are two bounded ways to explain evidence. They are not a universal animation style, and a URL in a brief does not make a claim true by itself.
Open the knowledge examples · Download the free Skill (ZIP)
Download and run a complete example
The package includes examples/wright-telegram.json and examples/phone-materials.json as complete source briefs. The complete ZIPs include the editable project, captions, source notes and local narration assets. Rendering creates a new video and cover:
unzip knowledge-explainer-wright-project-0.1.0.zip
cd knowledge-explainer-wright
npm run check
npm run preview
npm run render
Both ZIPs contain one top-level project directory. For the phone handoff, run unzip knowledge-explainer-phone-project-0.1.0.zip and then enter knowledge-explainer-phone/; do not add a destination directory that would nest the supplied project.
The current rendered examples include local English Kokoro narration from af_heart, captions and source notes. Narration remains source-dependent for new projects: narration text alone does not create audio. Supply checked local audio or use the package narration helper when the local Kokoro setup is available, then review the full picture and sound before delivery.
Create or revise a sourced project
Install the Skill from the ZIP or with the FeedSkills CLI tarball:
npx --yes --package=https://feedskills.com/downloads/feedskills-0.3.6.tgz feedskills install knowledge-explainer --agent codex
Run from your project root, outside the installed Skill directory, and create a fresh output folder outside .agents/skills/knowledge-explainer/:
node .agents/skills/knowledge-explainer/scripts/create-explainer.mjs create .agents/skills/knowledge-explainer/examples/wright-telegram.json .agents/skills/knowledge-explainer/scripts/film.mjs --out ./knowledge-wright-v1
cd ./knowledge-wright-v1
npm run check
npm run preview
npm run render
Every scene carries claimIds, every factual or contested claim carries sourceIds, and each source has an ID, title, HTTPS URL and note. Keep rights explicit for original artwork or source material. Preserve stable scene IDs and make a new output folder for revisions; a changed scene ID represents a new scene.
Sources used by the bundled briefs
The Wright example compares the 57-second telegram transcription with the 59-second Library of Congress record. It states the two-second difference and leaves the cause open. Check the source pages before publication: National Park Service — Wright brothers and Library of Congress — The Dream of Flight.
The phone example uses a schematic to distinguish three material roles: silicon computing, copper conduction and lithium-ion energy storage. Check the current source wording: Royal Society of Chemistry — Silicon, Royal Society of Chemistry — Copper, U.S. Department of Energy — How Batteries Work and U.S. Department of Energy — Lithium-ion batteries.
Scope and cost
The Skill is MIT-licensed and uses the pinned free HyperFrames 0.8.181 runtime. It includes no external archive image, sticker, music, logo or font, and it does not call a paid generation API. The first local render may download HyperFrames and a browser; optional local narration may download model data. Agent-provider fees, source verification and rights review remain separate from the local export.