ai-news-video
Skill2 starsChanged 2 months agoMITView on GitHub
Turn an AI model/product announcement into a short, voiced YouTube news video in Andy's voice. Use when Andy says "make a news video about <url>", "Anthropic/OpenAI/Google just shipped X", "turn this announcement into a video", or wants to repurpose an Andy's Take blog post into video. Orchestrates: fetch source -> draft script (Andy's voice) -> fact-check sheet -> [Andy approves] -> ElevenLabs voiceover -> Higgsfield B-roll -> FFmpeg assembly -> [Andy previews] -> YouTube metadata. Two human checkpoints. Speed-first: target announcement-to-publish-ready under 2 hours.
In this repository9 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Before drafting anything, read the project's voice docs and apply them:
- Very specific: line 48
Write two outputs to output/<slug>/:
- Very specific: line 51
Write visual-and-metadata.md — Higgsfield prompt list, real-capture shot list, fact-check sheet, and YouTube title/description/tags.
- Very specific: line 61
B-roll: run the higgsfield generate create ...
- Very specific: line 82
Keep output/<slug>/source.txt and the fact-check sheet after publishing — they're your record if a take needs correcting later.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AI News Video
Produces a short news video (target 2–3 min) from a single source URL, in Andy's voice, fast. This channel competes on speed and a sharp marketer angle, not polish. First-to-publish with a genuinely useful take wins the search traffic in the first 24 hours after a release.
Voice & accuracy are non-negotiable
Before drafting anything, read the project's voice docs and apply them:
- Speaking Style Guide — how Andy talks (the words, rhythm, "Right.", "I think", "in a nutshell", "cheers").
- Kallaway Viral Scripting Engine — structure (hook format, 2-1-3 body order, V-A-V intro, native CTA).
- Tone of Voice Guide — the "explain the consequence, not the engineering" reframing rule.
- Who Is Andy — context and the family/personal guardrails.
Structure comes from Kallaway. Words come from the Speaking Style Guide. Run the "Does this sound like Andy?" test before showing him the script.
Accuracy rule: This is a news channel built on trust. Every specific claim (numbers, "X times", prices, "available on all plans") must be traceable to the source. Produce a fact-check sheet alongside every script. Flag any claim whose scope is narrower than the script implies (e.g. a stat that's about code specifically, not "everything"). Never let an AI model generate a fake product screenshot, fake benchmark, or fake UI — real product shots are screen captures Andy records.
The pipeline
Stage 1 — Fetch & understand the source
- Fetch the URL (WebFetch, or
scripts/fetch_article.py <url> > output/<slug>/source.txtfor a saved copy). - For breaking releases, also pull one or two secondary sources (TechCrunch / VentureBeat) to cross-check facts, but the primary source is the source of truth.
- Identify the marketer angle: what does this change for a marketer or solo creator on Monday morning? Most coverage leads with benchmarks/engineering. Andy's job is the reframe. Lead with it.
Stage 2 — Draft the script
- Pick a Kallaway hook format (Contrarian and Investigator usually fit Andy best).
- Order the body 2-1-3 by impact.
- Write two outputs to
output/<slug>/:script-master.md— 2-column table (visual directions + spoken script).script-teleprompter.txt— clean spoken-only, ready for ElevenLabs. ~150 wpm; 350–450 words ≈ 2.5 min.
- Write
visual-and-metadata.md— Higgsfield prompt list, real-capture shot list, fact-check sheet, and YouTube title/description/tags.
Stage 3 — CHECKPOINT 1 (Andy approves the script)
Show Andy the teleprompter script + the fact-check sheet. He verifies claims, edits voice, approves. Do not render voiceover until he approves. This is the gate that protects the brand.
Stage 4 — Generate assets (run in parallel)
Kick these off together the moment the script is approved:
- Voiceover:
scripts/voiceover.sh output/<slug>/script-teleprompter.txt output/<slug>/voiceover.mp3 - B-roll: run the
higgsfield generate create ...commands fromvisual-and-metadata.md. Save clips tooutput/<slug>/clips/. (Uses the installedhiggsfield-generateskill / CLI.) - Remind Andy to record the real screen captures from the shot list in parallel.
Stage 5 — Assemble
scripts/assemble.sh output/<slug> stitches voiceover + clips + captions + intro/outro into
output/<slug>/final.mp4. Assembly is the part most likely to need hand-tuning — for v1, Andy may
prefer to assemble in his editor (CapCut/Resolve) while everything else stays automated. That's fine.
Stage 6 — CHECKPOINT 2 (Andy previews)
Andy watches final.mp4 once. Confirms title/description/tags from visual-and-metadata.md.
Then he uploads manually. Do not auto-publish news content — there's no taking back a bad take.
Repurposing an "Andy's Take" blog post
Same pipeline, simpler: skip Stage 1 monitoring, feed the published article as the source, allow a longer runtime (5–8 min), no fact-check sheet needed (it's Andy's own thinking). The reframe step becomes "tighten to the single sharpest argument for video."
Notes
- Avatar intro (HeyGen) is deliberately out of scope for v1: it adds render latency and a failure point that breaks the speed target. Add it later as an optional opener for evergreen videos.
- Keep
output/<slug>/source.txtand the fact-check sheet after publishing — they're your record if a take needs correcting later.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

