YapCut by alexmuresan.com

A measured-first short-form content engine for Claude Code. Its job is to make you the person your audience thinks of for one subject, and it is judged on what gets posted and answered, never on how many scripts it writes. Creator-agnostic: on first use it interviews you about who should recognise you and for what, what only you can say, and how you actually talk, then works to that.

1. Outlier Radar researches your niche every week: it finds short-form videos that beat their own creator's baseline by 5x or more (a proven mechanic, not a follower count), extracts the hook and format, pours in substance from your subject and your proof, and writes scripts and LinkedIn posts to the filming slots you actually have, on a tracked dashboard. Each week starts by reading last week's numbers.
2. Lead Magnet Report turns one category's measured data into a comment-gated document post: the finding free in the feed, the raw sheet behind a comment.
3. TikTok Yap Editor turns your raw phone clips into a finished, captioned, gated vertical video, talking-head or day-in-the-life, with an edit record that joins the cut to the post.

Pick a script, shoot it, drop the clips on the editor, paste your numbers back. That is the loop.

Install (Claude Code)

/plugin marketplace add alexdailycheckin/yapcut
/plugin install outlier-radar@yapcut
/plugin install tiktok-yap-editor@yapcut

Then say run outlier radar to start onboarding. Loading a skill does not start it. There is one install route on purpose: a copy under ~/.claude/skills/ shadows any plugin install silently and never updates, so this README no longer offers it.

Or paste this into Claude Code and let it do the setup and verify it:

Install YapCut (Alex Mureșan's short-form content engine) into my Claude Code setup and
verify it runs. Do the work, don't just print instructions.

1. Add the marketplace and install both plugins:
   /plugin marketplace add alexdailycheckin/yapcut
   /plugin install outlier-radar@yapcut
   /plugin install tiktok-yap-editor@yapcut
   Find the installed editor's path in ~/.claude/plugins/installed_plugins.json.

2. Install the editor's dependencies (macOS + Homebrew): run its scripts/setup_fonts.sh,
   then: brew install whisper-cpp; mkdir -p ~/.whisper-models; download
   https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.en.bin into it;
   pip3 install -r <repo>/requirements.txt (Pillow).

3. Verify: python3 <editor>/scripts/preflight.py must print ALL GOOD. Apply its fix hints and
   re-run until green. If something cannot be fixed, say which check failed and what still
   works without it.

4. Tell me to run /reload-plugins myself, then end with the exact sentence I type next:

   run outlier radar

   First run interviews me (who should recognise me and for what, what only I can say, where
   I already talk unscripted), creates the workspace at ~/outlier-radar/, writes the first
   real batch to my filming slots, sets up the weekly run, and opens the dashboard.

The research side works anywhere Claude Code runs. The editor is macOS + Homebrew (fonts, libass ffmpeg, whisper.cpp); on Linux or Windows you get Outlier Radar and the editor will not pass preflight.

What you get, honestly, by day 7

• A workspace (~/outlier-radar/) that is yours: radar-config.json with your audience, proof, quantity and voice settings, positioning.md, methods.md, leaders-to-study.md, your voice corpus, your weeks, your performance ledger. Plugin updates never touch it.
• One batch written to your filming slots (default 3 video slots plus 2 spares and 5 LinkedIn slots for a new creator), each script gated by machine (radar_gate.py: schema, voice, hooks, sources, visuals) before you see it, on a dashboard that opens on the scripts (pick a lane, read, film), with comment ammo for your daily comments, a proof chip on every card, and this week's one measured question.
• A weekly rhythm that starts by pasting your analytics table (five minutes) and reads the result before writing anything. Skip the paste and the digest says so and the batch ships.
• Not yet, by design: your voice will be thin until you have harvested 20 to 30 minutes of yourself talking unscripted (calls, voice notes, a podcast); methods.md holds the plays only you can supply; the secondary "viral" lane is off for the first four weeks.

What it needs from you in week one

Ask	Time	Why
One sentence: who should recognise you, and for what	2 min	every gate, the selector and the north star read it
Five things you have done a competitor cannot claim, three plays you run, ten lines you would never say	20 min	the un-copyable material; adjectives about your voice produce pastiche
Transcripts of you talking unscripted (or a 20 to 30 minute phone recording)	0 to 30 min	the voice is measured, never described
How many videos you will actually film this week	1 min	scripts you do not film teach the loop nothing
Paste your analytics table once a week	5 min	the loop

Use

• Ideas: "run outlier radar". Weekly by schedule or on demand. "hot drop" for a same-day post when something breaks in your niche.
• Numbers: paste the LinkedIn activity table or the TikTok Studio table when asked; ingest_feed.py turns it into rows, log_perf.py --report reads it back and writes performance/last-report.md.
• LinkedIn: select_linkedin.py picks each post's shape, job and posting day from its substance and from what your own numbers have learned.
• Distribution: the daily block (comments with receipts, connection requests, the reply block), day-0 posts, the people ledger, callback and promise: references/distribution.md.
• Edit: drop a folder of clips and say "make a tiktok from these" (Mode A) or "script my voiceover and cut my day clips" (Mode B). finalize.sh writes the edit record that joins the cut to the post.

What's inside

• plugins/outlier-radar/skills/outlier-radar/: the playbook (SKILL.md, about 260 lines with a routing table), the gate runner and gates, the ledger scripts, build_dashboard.py plus dashboard/, carousel.py, references/ (schema, receipts, distribution, hooks, LinkedIn craft, the show template), radar-config.example.json, a bundled example week.
• plugins/outlier-radar/skills/lead-magnet-report/: the report builder, the request ledger, the measurement law.
• plugins/tiktok-yap-editor/skills/tiktok-yap-editor/: the editor playbook, the gate ladder (scripts/gates.sh), scripts/yaplib/, the drivers, brand-config.example.json, and scripts/hfkit/, which packages a finished cut in HyperFrames (renders with npx hyperframes, no account; the curated sounds download with a free HeyGen account, or fall back to a CC0 pack).
• rules.json in each plugin: the numbers and rules a second surface has to apply the same way (cut rules, caption presets, hook limits, gate thresholds, the week schema, the script patterns). The scripts read them instead of carrying copies, and a phone editor reads the same files, so a rule changes in one place.
• tests/: a synthetic fixture that drives the editor end to end, the radar gates on the example week, and rules_check.py, which fails when a rule is copied back into a script. release.sh is the one release command.

What's new in 3.24.2

The audit release. The engine used to optimise the one stage that was never the constraint: 157 scripts written, 22 posted, the loop run once. 3.5.0 asks who your audience is before it asks your niche, writes to filming slots instead of a script bar, ships a gate that actually gates, closes the loop on disk (tracking, a people ledger, learned weights, one measured question a week), and ships the distribution doctrine. Full notes for this and every earlier release: CHANGELOG.md.

Author

Created by Alex Mureșan (alexmuresan.com). If this is useful, a credit and a link back are appreciated.

License

Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) (full text in LICENSE; attribution in NOTICE). Free to use, modify and share for noncommercial purposes with credit to "Alex Mureșan (https://alexmuresan.com)". Commercial use, including running it inside a for-profit company's operations, needs a separate license: [email protected].