Skip to content
Back to search

i-hate-editing

Drop in raw talking-head takes, get back a finished, publishable video. Every rule in this repo was paid for by a rejected render.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/ranahaani--i-hate-editing/badge.svg)](/listings/ranahaani--i-hate-editing)

HTML

<a href="/listings/ranahaani--i-hate-editing"><img src="/api/listings/ranahaani--i-hate-editing/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Option A — skills.sh

npx skills add ranahaani/i-hate-editing

That registers the skill with your agent. You still need the local tools (`ffmpeg`, `whisper-cpp`, Playwright Chromium). Details in [`install.md`](./install.md).

Option B — paste this into your agent

Works in Claude Code, Cursor, Codex, and anything else with a shell:

Set up https://github.com/ranahaani/i-hate-editing for me.

Read install.md first: clone the repo to a stable path, symlink it into this
agent's skills directory, install ffmpeg + whisper.cpp, create the Python venv
with playwright + pillow, and optionally run scripts/sfx_library.py install.
Then read SKILL.md for daily usage. After install, don't start editing —
tell me it's ready and wait for me to drop footage into a folder.

Option C — by hand

git clone https://github.com/ranahaani/i-hate-editing ~/Developer/i-hate-editing
ln -sfn ~/Developer/i-hate-editing ~/.claude/skills/i-hate-editing   # Claude Code
ln -sfn ~/Developer/i-hate-editing ~/.cursor/skills/i-hate-editing   # Cursor
# ln -sfn ~/Developer/i-hate-editing ~/.codex/skills/i-hate-editing  # Codex

cd ~/Developer/i-hate-editing
uv venv .venv && uv pip install --python .venv playwright pillow
.venv/bin/playwright install chromium
brew install ffmpeg whisper-cpp   # macOS; use your package manager on Linux

First edit

cd /path/to/your/footage
claude   # or your agent of choice

> edit these into a reel

It scans the machine, asks five questions, picks a Whisper model for your language and hardware, and starts. More in [`install.md`](./install.md).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 437 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. ---
  2. name: i-hate-editing
  3. description: Turn raw talking-head takes into a finished, publishable video package — cut, captioned, scored, graded, with platform variants, post caption and thumbnail. For short-form vertical and YouTube talking-head footage. Opinionated by design; the studio makes the craft decisions and takes plain-language direction. Use when someone drops recorded footage in a folder and wants it edited.
  4. ---
  5. # I Hate Editing
  6. A studio, not a tool. Someone hands you raw takes; you hand back something they
  7. can publish without touching an editor.
  8. **The bar:** if a professional editor returned this, would the client accept it?
  9. That question decides every call you make. "The pipeline ran" is not done.
  10. ## Principles
  11. 1. **Judgement is the product; configuration is a failure state.** Never ask
  12. which font, which transition, which decibel level. Decide, show the result,
  13. and take plain direction — *punchier*, *slower*, *less music*. The user
  14. describes the feeling; you own the numbers.
  15. 2. **The rules are not suggestions.** `rules/` holds craft decisions that were
  16. each paid for by a rejected render, and `HARD-RULES.md` holds things that
  17. fail silently if you deviate. Read both before editing. Artistic freedom
  18. without taste memory produces the median boring cut every time — that is
  19. the failure this skill exists to prevent.
  20. 3. **Audio is the authority on time.** Cut boundaries come from silence
  21. detection and word onsets, never from eyeballing video.
  22. 4. **Verify before you present.** If you did not look at the frame or measure
  23. the level, it is not done. See "Verify" below — it is a gate, not a habit.
  24. 5. **Deliver a package.** A cut file is not the deliverable. See "Output".
  25. 6. **Learn from every correction.** Feedback becomes a durable rule in
  26. `taste.md`, read before every future edit. The studio should be better on
  27. the tenth video than the first.
  28. 7. **Never fabricate proof.** When footage references a real repo, page, tweet
  29. or product, capture the real thing. Never mock up something that implies a
  30. screenshot of reality.
  31. ## Setup
  32. First run performs a capability scan and asks what the user is making, then
  33. installs only what is missing:
  34. ```bash
  35. python3 scripts/scan.py
  36. ```
  37. It writes `profile.yml` (brand, language, pacing, caption style, aspect
  38. targets) and scaffolds the project. Everything downstream reads that file —
  39. there are no hardcoded brand values anywhere in this skill.
  40. Requirements, all free:
  41. | Job | Tool | Notes |
  42. |---|---|---|
  43. | Transcription | `whisper.cpp` | Local. Model chosen by language + hardware. |
  44. | Media processing | `ffmpeg` / `ffprobe` | |
  45. | Captions, cards, compositing | HyperFrames | Renders HTML/CSS/GSAP compositions. |
  46. | Designed scenes | Remotion | React motion graphics, one composition per scene. Installed per project by `scenes.py init`. |
  47. | Composition runtime | `node` 22+ | Already required by Claude Code. |
  48. | B-roll download | `yt-dlp` | Optional. |
  49. Nothing requires an API key and nothing leaves the machine.
  50. ## Project layout
  51. Footage lives wherever the user put it. All output goes in `<footage>/studio/`.
  52. ```
  53. <footage>/
  54. ├── <source takes, untouched>
  55. └── studio/
  56. ├── profile.yml brand, language, pacing — the only config
  57. ├── taste.md durable feedback; read before every edit
  58. ├── transcripts/ cached per source, never re-transcribed
  59. ├── takes.md packed phrase-level transcript (reading view)
  60. ├── edl.json cut decisions
  61. ├── composition/ HyperFrames project
  62. ├── scenes.json designed scenes: which beat, what moves
  63. ├── scenes/ Remotion project for those scenes
  64. ├── assets/ fetched SFX, logos, screenshots
  65. ├── verify/ frames and measurements from the verify pass
  66. └── out/ the delivered package
  67. ```
  68. ## The process
  69. Run in order. Two stop gates; do not run past them.
  70. **1 — Read state.** Load the taste memory and profile before anything else:
  71. ```bash
  72. python3 scripts/taste.py --studio <footage>/studio brief
  73. ```
  74. Those instructions override the defaults in `rules/` — they are what this
  75. particular person has already corrected you on. Never regenerate an artifact
  76. that is already on disk.
  77. **2 — Inventory.** `ffprobe` every source. Note resolution, fps, duration,
  78. audio channels. Flag anything that will bite later (variable frame rate,
  79. mismatched fps between takes, silent audio track).
  80. **3 — Transcribe.** `scripts/transcribe.py <sources>` — word-level, cached,
  81. language pinned from `profile.yml`. Never auto-detect: auto-detect silently
  82. *translates* some languages, which corrupts every downstream timestamp.
  83. **4 — Pack.** `scripts/pack.py` produces `takes.md`, the phrase-level reading
  84. view. This is what you actually read to make cut decisions — not raw JSON,
  85. and not the video.
  86. **5 — Plan the cut.** Identify beats, pick the best take of each, and place
  87. boundaries. This is judgement work; `rules/cutting.md` has the craft.
  88. Retakes cluster at file boundaries — expect one wherever a source ends.
  89. **6 — Assemble.** Write `edl.json`, then `scripts/render.py` for a draft.
  90. **7 — Verify the cut.** *Gate.* See below. Fix and re-verify before showing
  91. anything.
  92. **8 — Present the cut.**
  93. ```bash
  94. python3 scripts/review.py --studio <footage>/studio
  95. # phone preview on a trusted network only:
  96. # python3 scripts/review.py --studio <footage>/studio --lan
  97. ```
  98. Opens it in a browser with frame-stepping on **localhost**. Pass `--lan` only
  99. on a trusted network if you need a phone URL — that mode has no auth, and a
  100. machine's IP changes with the network anyway. **STOP.** Do not build captions,
  101. motion or sound on an unapproved cut — every downstream timestamp depends on
  102. it, and a base-cut change invalidates all of them.
  103. **9 — Enhance the voice.** Every reel, right after the cut is approved and
  104. before any sound effect is placed:
  105. ```bash
  106. python3 scripts/enhance.py <studio>/cut.wav -o <studio>/voice_enhanced.wav
  107. ```
  108. It runs Adobe Podcast Enhance Speech v2 through the signed-in ego-browser
  109. session (upload, poll, download the stems) and mixes the website's default
  110. "Speech 50%, Background 10%, Music 10%" locally. About 30s for a short reel.
  111. Then:
  112. 1. Level it: `acompressor=threshold=0.045:ratio=3.2:attack=6:release=110:makeup=1`,
  113. then gain and `alimiter=limit=0.7` to land near -18.4 LUFS. Boosting it plain
  114. to -15 LUFS pushes speech into the limiter and every sting collapses.
  115. 2. Remux it as the cut's audio (`-c:v copy`), so no video re-render is needed and
  116. every timestamp holds. The script fails if the duration drifts over 50 ms.
  117. 3. Place stings and mix against this enhanced voice, not the raw take.
  118. If it prints `NOT_SIGNED_IN`, ask the user to sign in to podcast.adobe.com in
  119. ego-browser. Never type credentials. Use `--mix full` only when asked; 100%
  120. enhanced speech sounds processed.
  121. **10 — Enrich.** In order:
  122. ```bash
  123. python3 scripts/grade.py cut.mp4 --strength normal # look at the comparison
  124. python3 scripts/captions.py --studio <studio>
  125. python3 scripts/proofread.py fix --studio <studio> --name "Claude" --name "<tool>"
  126. python3 scripts/capture.py <url> --studio <studio> --find "<the phrase>"
  127. python3 scripts/icons.py fetch claude github --studio <studio>
  128. # Generated stills for beats with nothing real to show — see "Generated stills".
  129. python3 scripts/gen_image.py "<full prompt>" --studio <studio> --name payoff --animate 3
  130. # Optional: 2–3 AI B-roll clips (3s each) — writes studio/broll.json + assets/broll/
  131. # Author studio/slots.json from takes.md beats first; see broll-gen skill.
  132. ~/.claude/skills/broll-gen/scripts/broll-gen generate --studio <studio> --slots <studio>/slots.json --duration 3
  133. python3 scripts/sfx.py plan --studio <studio> --library <sfx>
  134. # Designed scenes for beats with structure — see "Designed scenes".
  135. python3 scripts/scenes.py init --studio <studio>
  136. python3 scripts/scenes.py check --studio <studio>
  137. python3 scripts/scenes.py render --studio <studio>
  138. python3 scripts/compose.py --studio <studio> --render
  139. ```
  140. ### Generated stills
  141. `scripts/gen_image.py` talks to a local Gemini proxy
  142. (`http://127.0.0.1:8081/v1`, key `sk-gemini`) — no account, no cost, no
  143. watermark. Image generation there is a **route, not a model**: nothing in
  144. `/v1/models` makes images, `n` must be 1, and `size`/`quality`/`style` are
  145. rejected, so the aspect goes in the prompt text. The script injects the
  146. vertical-9:16 clause for you and warns if what comes back is landscape. It
  147. injects nothing else — when a beat needs text, logos, screens or UI kept out,
  148. put that ban at the end of that beat's own prompt.
  149. **Reach for it only after checking there is nothing real to show.** A repo, a
  150. page, a dashboard always beats a generated still — see `rules/proof.md`. And a
  151. generated image must never imply a screenshot of something real.
  152. **Write the whole prompt.** A short prompt returns a stock-photo cliché. Every
  153. prompt states, in this order: the subject, the setting, the camera (lens,
  154. aperture, height), the lighting with its direction and colour temperature, the
  155. composition including which parts of the frame stay empty, the mood, and the
  156. ban list. `--animate N` renders a push-in clip from the still, because a
  157. generated still that sits motionless is a dead frame; add the clip to
  158. `broll.json` as an overlay.
  159. Worked prompt, for the "gold mine" payoff beat of a reel about saving money:
  160. > A deep underground cavern of glowing golden light, with streams of tiny
  161. > luminous golden particles rising out of the darkness like sparks. Rich black
  162. > background, warm gold highlights, high contrast, dramatic rim lighting,
  163. > shot on a 35mm lens at f2, abstract and atmospheric, the lower third almost
  164. > pure black so captions stay readable. Cinematic, fine film grain.
  165. Full prompt library and failure log: `~/ai-me/reels/docs/agent-suggestions.md`
  166. and `~/.claude/skills/virtual-bg/references/plate-prompts.md`.
  167. **Do not generate a virtual background.** Reels ship with the real room — see
  168. `taste.md` under framing, and the STOP notice in the `virtual-bg` skill.
  169. You author four files by hand, the same way you author the EDL. They are the
  170. edit; the scripts only render them.
  171. **`cards.json`** — the motion vocabulary. Without it the piece is a face with
  172. captions, and `beats.py` will fail step 11.
  173. ```json
  174. {"cards": [
  175. {"start": 0.35, "duration": 5.1, "style": "band", "big": "SIX WORDS OR FEWER"},
  176. {"start": 13.1, "duration": 4.2, "kicker": "small label above",
  177. "big": "HEADLINE WITH *ACCENT* WORD", "sub": "one supporting line",
  178. "items": [{"title": "row one", "note": "right-aligned"},
  179. {"title": "row two", "note": "staggered in"}]},
  180. {"start": 21.6, "duration": 1.9, "full": true, "big": "OWNS THE *FRAME*"}
  181. ]}
  182. ```
  183. Add `"icon": "claude.svg"` to put a brand mark on a card, and
  184. `"brand_colour": "#D97757"` to theme the whole card in that brand's colour —
  185. `icons.py` reports the official hex when it fetches. A named tool with its own
  186. mark on its own colour reads as designed; the same dark card every time reads
  187. as a template.
  188. `style: "band"` is the hook headline over a full frame. Default is a
  189. half-screen split: card on top, face below. `full: true` takes the whole frame
  190. and hides the face — captions are suppressed under it. `*asterisks*` mark the
  191. accent word. `items` makes it a numbered list. Choose the form from what the
  192. sentence is doing, and the arrival is chosen for you — see `rules/motion.md`.
  193. Captions come out of `captions.py` as a raw pass and are marked
  194. `proofread: false`. `proofread.py` restores product names and strips
  195. punctuation artefacts, but it cannot fix meaning — so it only marks them
  196. proofread when a model pass ran (`--llm "<any command>"`) or when you rewrote
  197. the copy yourself and passed `--accept`. Verification fails while the flag is
  198. false, because captions are the most-read thing on screen.
  199. **`proof.json`** — real pages: which capture, which target, which move.
  200. ```json
  201. {"beats": [
  202. {"asset": "repo", "start": 6.9, "duration": 3.1,
  203. "action": "scroll", "from_y": 60, "to_y": 560},
  204. {"asset": "repo", "start": 10.1, "duration": 2.9, "action": "zoom",
  205. "target": "the exact text", "highlight": true, "highlight_at": 0.75}
  206. ]}
  207. ```
  208. `asset` is the `--name` you gave `capture.py`. `target` must be text that
  209. capture found. `rules/proof.md` has the craft.
  210. **`scenes.json`** — designed scenes, built in Remotion. A card is text on a
  211. ground; a scene is a built visual with structure — a counter running, a
  212. comparison assembling, a stylised UI typing. Beats with structure go here
  213. rather than into a headline and a sub-line.
  214. ```json
  215. {"scenes": [
  216. {"id": "token-waste", "start": 12.4, "duration": 3.2, "full": true,
  217. "line": "every retry re-sends the whole conversation",
  218. "visual": "context bar fills, three retry chips stack on it, bar turns red on the third",
  219. "assets": ["claude.svg"]}
  220. ]}
  221. ```
  222. `scenes.py sync` writes the Remotion project's `Root.tsx` and
  223. `src/data/scene-spec.json` from this file and the cut — one composition per
  224. scene, at the cut's own size and frame rate. You write only the components, in
  225. `<studio>/scenes/src/scenes/<Component>.tsx`. `scenes.py studio` opens Remotion
  226. Studio to preview them, `render` writes one MP4 per scene and registers them as
  227. overlays so `compose.py` composites them under the captions.
  228. Show `scenes.json` before writing any component — a component written against
  229. an unapproved beat is the expensive artefact to throw away. Craft and the
  230. self-review checklist: `rules/scenes.md`. Scene stings go in `sfx.json`, never
  231. inside the component: overlays composite muted (`HARD-RULES.md`).
  232. **`profile.yml`** — written by `scan.py`, or by hand when it cannot run
  233. interactively:
  234. ```yaml
  235. language: ur
  236. translate_captions: true
  237. aspects: ["9:16"]
  238. pacing: punchy # punchy | balanced | restrained
  239. brand: {accent: "#76B900", font: "Archivo Black"}
  240. transcription: {model: large-v3-turbo}
  241. face_half_y: 30 # vertical crop of the face in a split
  242. ```
  243. This is where a trimmed recording becomes an edited video.
  244. **11 — Verify the render.** *Gate.* Frames and levels, again, plus the beat
  245. map:
  246. ```bash
  247. python3 scripts/beats.py --studio <studio> # stretches with nothing new
  248. python3 scripts/sound.py check out/master.mp4 # stings audible, not just present
  249. ```
  250. A shot that sits still is the most-reported defect in short-form and the
  251. easiest to miss, because nothing errors when a face holds for ten seconds.
  252. **12 — Deliver the package.**
  253. ```bash
  254. python3 scripts/music.py out/master.mp4 <bed>.mp3 -o out/final.mp4
  255. python3 scripts/deliver.py --studio <studio>
  256. python3 scripts/review.py --studio <studio>
  257. ```
  258. Then write `out/post.md` yourself from the spoken lines it surfaces, and look
  259. at the thumbnail candidates and pick one. Neither is generated: a generated
  260. title reads like a generated title, and "face clear, eyes open" is not a
  261. metric.
  262. **13 — Learn.** Record every correction the user made, phrased as an
  263. instruction for next time and filed under the area it affects:
  264. ```bash
  265. python3 scripts/taste.py --studio <footage>/studio add captions \
  266. "Start each caption 0.08s after the word is spoken, never before" \
  267. --said "we show early for a sec then I start speaking"
  268. ```
  269. Always pass `--said` with their actual words. The instruction is your reading
  270. of the feedback and can be wrong; keeping the original means a bad reading can
  271. be corrected later instead of quietly hardening into a rule.
  272. When a rule stops applying, retire it rather than deleting it —
  273. `taste.py retire <area> <n>` — so the reversal stays visible.
  274. ## Verify
  275. Two failures survive every automated check and both get flagged by users, so
  276. hunt them explicitly.
  277. **Repeated phrases at seams.** A join can leave a word said twice. Transcribe
  278. the rendered cut in short windows aligned to each seam — never one whole-file
  279. pass, which condenses and hides the defect.
  280. **Clipped clauses.** Every kept block must start and end on a complete
  281. thought. A cut that lops a negation inverts the meaning of the sentence. If
  282. the clause only completes in a later take, stitch the two rather than shipping
  283. the fragment.
  284. Then: `ffprobe` the duration against what the EDL predicts, extract frames at
  285. every boundary and look at them, and measure audio levels at each sound effect
  286. against a voice-only baseline. "Not clipping" is not the same as "audible" —
  287. see `rules/sound.md`.
  288. ## Output
  289. Every finished run delivers a package, because that is what an editor hands
  290. back:
  291. - **Master** — graded, mixed, captioned
  292. - **Platform variants** — vertical, square, landscape as `profile.yml` requests
  293. - **Post caption** — hook line, substance, hashtags
  294. - **Thumbnail frame** — pulled from the cut, face visible, no mid-blink
  295. - **Title options** — a small set to choose from
  296. ## House style
  297. Locked on the NVIDIA and Promptive reels (2026-10-02). Apply it by default; the
  298. details and the reasons are in the rules files named.
  299. - **Hook, 0 to about 3.4s** (`rules/hooks.md`, "The split-proof hook"):
  300. - Split from frame 0, with an HD visual on top and the face below.
  301. - Text: 2-3 familiar words plus one emoji ("CLAUDE / LIMIT ISSUE 😕").
  302. - The top panel changes once at 1.0-1.3s.
  303. - Named tools' marks pop in on their spoken names.
  304. - A loss word turns the whole frame red for about 0.5s.
  305. - Plan it with `hook_plan.py --proof <capture>`.
  306. - **Everything HD:**
  307. - The hook's top panel is a motion graphic built at frame resolution, never
  308. an upscaled screenshot.
  309. - Rebuild product screenshots as native Flick/Remotion scenes at 1080x960,
  310. using the real UI's text (`rules/proof.md`).
  311. - **Palette:** green `#76B900` for emphasis words in captions, accents, chips
  312. and borders. The ground is near-black green with a slow-drifting grid. It is
  313. the profile default; a brand colour in `profile.yml` still wins.
  314. - **Sound:**
  315. - Pop sting: `assets/sfx/pop_dragon.mp3` (peak 0.197s; start = hit - 0.197).
  316. - Impact on frame 1, swoosh on a panel swap, click on a highlight.
  317. - Every sting +2 to +4 dB above the enhanced voice (centre +3.1). Master at
  318. -15 LUFS.
  319. - **Voice:** Adobe Enhance (step 9) on every reel.
  320. - **Cut:** transcribe the first 0.3s of every block on its own. A clipped false
  321. start ("com" before "link chahiye") hides at block heads.
  322. ## Craft rules
  323. Read the relevant file before touching that part of the edit. Each rule states
  324. the failure it prevents.
  325. | File | Covers |
  326. |---|---|
  327. | `HARD-RULES.md` | Correctness. Silent failures. Non-negotiable. |
  328. | `rules/cutting.md` | Take selection, boundaries, silence, pacing |
  329. | `rules/hooks.md` | Openings, headline, retention |
  330. | `rules/captions.md` | Timing, chunking, style |
  331. | `rules/sound.md` | Sound effect placement, levels, music |
  332. | `rules/proof.md` | Screenshots, B-roll, zoom and highlight |
  333. | `rules/motion.md` | Zooms, transitions, cards, layout |
  334. | `rules/scenes.md` | Designed Remotion scenes: when, how, what to check |
  335. | `rules/framing.md` | Crops, splits, composition |
  336. | `rules/assets.md` | Where pictures come from: proof, reference clips, flick, generated stills, Flow |
  337. ## Anti-patterns
  338. - Asking the user to configure something you should have decided.
  339. - Building captions or motion before the cut is approved.
  340. - Trusting a whole-file transcript to verify a cut.
  341. - Concluding a sound effect works because it did not clip.
  342. - Re-transcribing a source that has not changed.
  343. - Auto-detecting the spoken language.
  344. - Presenting output you did not look at.
  345. - Rendering a sentence with structure as a headline and a sub-line.
  346. - Writing a scene component before `scenes.json` was agreed.
  347. - Baking a sound effect into a scene, where the mix never hears it.
  348. - Adding a feature nobody asked for.
  349. - Finishing an edit without recording what the user corrected.
README.md · README · 329 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # I Hate Editing
  2. [![Install](https://img.shields.io/badge/skills.sh-npx%20skills%20add-111111)](https://github.com/ranahaani/i-hate-editing#option-a--skillssh)
  3. [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](./LICENSE)
  4. [![Local](https://img.shields.io/badge/transcription-local%20whisper.cpp-brightgreen)](./install.md)
  5. [![No API key](https://img.shields.io/badge/cloud%20STT-not%20required-lightgrey)](./SECURITY.md)
  6. Drop in raw talking-head takes. Get back something you can publish.
  7. I named it after a note to myself: never hand back work that still needs fixing
  8. by hand.
  9. <p align="center">
  10. <img src="docs/demo.gif" alt="The same reel twice: the plain face-only cut on the left, the finished reel with proof cards, captions and sound on the right, pipeline log in a terminal below." width="820">
  11. </p>
  12. <p align="center">
  13. <em>Same reel on both phones, frame for frame. Left is the cut with nothing
  14. added; right is what shipped.
  15. <a href="https://www.instagram.com/ranahaani.me/reels/">More reels on Instagram</a>.</em>
  16. </p>
  17. The skill picks the take, decides the repo and the docs page were worth
  18. capturing, where to zoom, which line gets the highlighter, what goes on each
  19. card, and where every sound lands. You say "punchier." It owns the numbers.
  20. ## Reels made with it
  21. <p align="center">
  22. <a href="docs/reels/strix.mp4"><img src="docs/reels/strix.jpg" alt="Strix reel" width="220"></a>
  23. </p>
  24. <p align="center">
  25. <em>Strix, straight out of <code>studio/out</code>. Click to play.
  26. More on <a href="https://www.instagram.com/ranahaani.me/reels/">Instagram</a>.</em>
  27. </p>
  28. ## What it does
  29. - Cuts from silence and word onsets, not from scrubbing the timeline
  30. - Re-transcribes a short window around every seam so repeated words and
  31. mid-clause cuts get caught before you see a preview
  32. - Writes captions that still read on a phone, with a proofread gate before
  33. they ship
  34. - Captures real pages as tall stills, then scrolls, zooms, and highlights the
  35. line you said (no mocked screenshots)
  36. - Places free-licensed SFX by intent and aligns peaks so a short window is
  37. actually audible
  38. - Builds beats with structure as real motion graphics — Remotion components,
  39. previewed in Studio, composited under the captions
  40. - Hands back a package: master, platform crops, ranked thumbnails, post lines
  41. - Remembers corrections in `taste.md` so the tenth video is better than the first
  42. - Runs transcription on your machine with `whisper.cpp`. No cloud STT key.
  43. Built for talking-head reels and YouTube explainers.
  44. ## Quick start
  45. ### Option A — skills.sh
  46. ```bash
  47. npx skills add ranahaani/i-hate-editing
  48. ```
  49. That registers the skill with your agent. You still need the local tools
  50. (`ffmpeg`, `whisper-cpp`, Playwright Chromium). Details in
  51. [`install.md`](./install.md).
  52. ### Option B — paste this into your agent
  53. Works in Claude Code, Cursor, Codex, and anything else with a shell:
  54. ```text
  55. Set up https://github.com/ranahaani/i-hate-editing for me.
  56. Read install.md first: clone the repo to a stable path, symlink it into this
  57. agent's skills directory, install ffmpeg + whisper.cpp, create the Python venv
  58. with playwright + pillow, and optionally run scripts/sfx_library.py install.
  59. Then read SKILL.md for daily usage. After install, don't start editing —
  60. tell me it's ready and wait for me to drop footage into a folder.
  61. ```
  62. ### Option C — by hand
  63. ```bash
  64. git clone https://github.com/ranahaani/i-hate-editing ~/Developer/i-hate-editing
  65. ln -sfn ~/Developer/i-hate-editing ~/.claude/skills/i-hate-editing # Claude Code
  66. ln -sfn ~/Developer/i-hate-editing ~/.cursor/skills/i-hate-editing # Cursor
  67. # ln -sfn ~/Developer/i-hate-editing ~/.codex/skills/i-hate-editing # Codex
  68. cd ~/Developer/i-hate-editing
  69. uv venv .venv && uv pip install --python .venv playwright pillow
  70. .venv/bin/playwright install chromium
  71. brew install ffmpeg whisper-cpp # macOS; use your package manager on Linux
  72. ```
  73. ### First edit
  74. ```bash
  75. cd /path/to/your/footage
  76. claude # or your agent of choice
  77. ```
  78. > edit these into a reel
  79. It scans the machine, asks five questions, picks a Whisper model for your
  80. language and hardware, and starts. More in [`install.md`](./install.md).
  81. ## How it works
  82. The model does not watch the video. Audio is the clock. Scripts do the
  83. mechanical work; the agent only makes taste calls.
  84. ```mermaid
  85. %%{init: {
  86. "theme": "base",
  87. "themeVariables": {
  88. "fontFamily": "ui-sans-serif, system-ui, sans-serif",
  89. "fontSize": "14px",
  90. "primaryTextColor": "#0f172a",
  91. "lineColor": "#94a3b8",
  92. "clusterBkg": "#ffffff00",
  93. "clusterBorder": "#e2e8f0",
  94. "edgeLabelBackground": "#ffffff",
  95. "tertiaryColor": "#f8fafc"
  96. },
  97. "flowchart": {
  98. "curve": "basis",
  99. "padding": 16,
  100. "nodeSpacing": 32,
  101. "rankSpacing": 40,
  102. "htmlLabels": true
  103. }
  104. }}%%
  105. flowchart TB
  106. RAW(["🎬 raw takes"]):::src
  107. PROF(["🎛️ profile.yml"]):::cfg
  108. subgraph A["🎧 1 · LISTEN"]
  109. direction LR
  110. SCAN(["🔍 scan"]):::listen
  111. ASR(["🗣️ whisper.cpp"]):::listen
  112. PACK(["📄 pack"]):::listen
  113. SIL(["🔇 silences"]):::listen
  114. TAKES[("takes.md")]:::artBlue
  115. GAPS[("gaps.json")]:::artBlue
  116. SCAN --> ASR --> PACK --> TAKES
  117. SCAN --> SIL --> GAPS
  118. end
  119. subgraph B["✂️ 2 · CUT · HARD GATE"]
  120. direction LR
  121. EDL[("edl.json")]:::artAmber
  122. RND(["🎞️ render"]):::cut
  123. VER{"🛡️ verify<br/>seam re-ASR"}:::gate
  124. GRD(["🎨 grade"]):::cut
  125. CUTV[("cut.mp4")]:::artAmber
  126. EDL --> RND --> CUTV --> VER
  127. VER -- "❌ fail" --> EDL
  128. VER -- "✅ pass" --> GRD
  129. end
  130. subgraph C["✨ 3 · ENRICH"]
  131. direction LR
  132. CAP(["💬 captions"]):::enrich
  133. PRF(["🌐 proof"]):::enrich
  134. SFX(["🔊 sfx + music"]):::enrich
  135. CMP(["🧩 compose"]):::enrich
  136. MASTER[("master.mp4")]:::artPink
  137. CAP --> CMP
  138. PRF --> CMP
  139. SFX --> CMP
  140. CMP --> MASTER
  141. end
  142. subgraph D["🚀 4 · SHIP"]
  143. direction LR
  144. DEL(["📦 deliver"]):::ship
  145. PKG[("studio/out")]:::artGreen
  146. REV(["👀 review"]):::ship
  147. DEL --> PKG --> REV
  148. end
  149. RAW --> SCAN
  150. PROF -.-> SCAN
  151. PROF -.-> CAP
  152. TAKES --> EDL
  153. GAPS --> EDL
  154. GRD --> CAP
  155. GRD --> PRF
  156. GRD --> SFX
  157. MASTER --> DEL
  158. classDef src fill:#0f172a,stroke:#facc15,stroke-width:3px,color:#fef08a
  159. classDef cfg fill:#fef3c7,stroke:#f59e0b,stroke-width:2px,color:#92400e
  160. classDef listen fill:#dbeafe,stroke:#2563eb,stroke-width:2px,color:#1e3a8a
  161. classDef cut fill:#ffedd5,stroke:#ea580c,stroke-width:2px,color:#9a3412
  162. classDef enrich fill:#f3e8ff,stroke:#9333ea,stroke-width:2px,color:#581c87
  163. classDef ship fill:#dcfce7,stroke:#16a34a,stroke-width:2px,color:#14532d
  164. classDef gate fill:#fee2e2,stroke:#dc2626,stroke-width:3px,color:#7f1d1d
  165. classDef artBlue fill:#eff6ff,stroke:#3b82f6,stroke-width:2px,color:#1d4ed8
  166. classDef artAmber fill:#fffbeb,stroke:#d97706,stroke-width:2px,color:#b45309
  167. classDef artPink fill:#fdf4ff,stroke:#c026d3,stroke-width:2px,color:#86198f
  168. classDef artGreen fill:#f0fdf4,stroke:#22c55e,stroke-width:2px,color:#15803d
  169. style A fill:#eff6ff,stroke:#3b82f6,stroke-width:2px,color:#1e40af
  170. style B fill:#fff7ed,stroke:#f97316,stroke-width:2px,color:#c2410c
  171. style C fill:#faf5ff,stroke:#a855f7,stroke-width:2px,color:#7e22ce
  172. style D fill:#f0fdf4,stroke:#22c55e,stroke-width:2px,color:#15803d
  173. linkStyle 8 stroke:#dc2626,stroke-width:2.5px
  174. linkStyle 9 stroke:#16a34a,stroke-width:2.5px
  175. ```
  176. `verify` blocks the cut. `review` is when you look at the package. If you change
  177. the cut after captions and sound are built, those timestamps are wrong. Start
  178. that layer over.
  179. ```
  180. footage/
  181. ├── take-01.mp4 your sources stay put
  182. └── studio/ everything the skill writes
  183. ├── profile.yml
  184. ├── taste.md
  185. ├── transcripts/
  186. ├── edl.json
  187. ├── composition/
  188. └── out/ what you publish from
  189. ```
  190. ## Why this exists
  191. Two things matter. The rest is plumbing.
  192. Joins love to leave a word sitting there twice. Cuts land mid-clause and flip
  193. what you meant. A whole-file transcript papers over both. So after the cut
  194. renders, verify re-transcribes short windows on each seam. Repeats are
  195. mechanical. Broken clauses get flagged for a human call.
  196. And when you correct something, it sticks. "Captions feel early" goes into
  197. `taste.md` and shows up on the next job. You should not have to teach the same
  198. lesson twice.
  199. You also should not be asked which font, which transition, or what decibel
  200. level. Say *punchier*, *slower*, *less music*. The studio owns the numbers.
  201. ## Proof B-roll
  202. Name a repo, an article, a number. The skill opens the real page in a vertical
  203. viewport, grabs one tall still, then scrolls, zooms, and draws a highlighter
  204. across the line you said.
  205. A screen recording locks scroll speed forever. A still plus motion in the
  206. composition can re-time with the cut and sit on the same timeline as your face,
  207. captions, and sound.
  208. Targets are text, not CSS selectors. If the line is not on the page, it says so
  209. instead of grabbing something nearby.
  210. ## The rules
  211. Each rule exists because a render failed once and we paid for it.
  212. | File | Covers |
  213. |---|---|
  214. | [`HARD-RULES.md`](./HARD-RULES.md) | Silent failures. Non-negotiable. |
  215. | [`rules/cutting.md`](./rules/cutting.md) | Take selection, seams, what to remove |
  216. | [`rules/hooks.md`](./rules/hooks.md) | Openings, headline, retention |
  217. | [`rules/captions.md`](./rules/captions.md) | Timing, chunking, style |
  218. | [`rules/sound.md`](./rules/sound.md) | Placement, levels, audibility |
  219. | [`rules/motion.md`](./rules/motion.md) | Zooms, pacing, easing |
  220. | [`rules/framing.md`](./rules/framing.md) | Crops, splits, composition |
  221. | [`rules/proof.md`](./rules/proof.md) | Screenshots, B-roll, zoom, highlight |
  222. Agent instructions: [`SKILL.md`](./SKILL.md). Threat model: [`SECURITY.md`](./SECURITY.md).
  223. ## Requirements
  224. | Tool | Why | Required |
  225. |---|---|---|
  226. | `ffmpeg` / `ffprobe` | Media | Yes |
  227. | `whisper.cpp` | Local transcription | Yes |
  228. | `node` 22+ | HyperFrames | Yes |
  229. | `playwright` + `pillow` | Proof capture, thumbnails | For proof / deliver |
  230. | `yt-dlp` | Reference footage | Optional |
  231. ## Limitations
  232. - Talking-head short-form and explainers. Not travel montages or multi-cam.
  233. - You bring the music bed. The skill ducks it; it does not license tracks.
  234. - Proof capture needs Chromium. Skip it if face + captions is enough.
  235. - First Whisper download is roughly 0.5–3 GB, depending on language and hardware.
  236. ## Contributing
  237. Best PR is a rule: what failed, what it cost, what you will never do again. Add
  238. it under [`rules/`](./rules/) or in [`HARD-RULES.md`](./HARD-RULES.md).
  239. ```bash
  240. uv venv .venv && uv pip install --python .venv pytest
  241. .venv/bin/pytest -q
  242. ```
  243. ## Disclaimers
  244. Footage stays on your machine. Scripts can still hit the network for proof
  245. pages, Mixkit SFX, a pinned HyperFrames package, and Remotion when a project
  246. uses designed scenes. See
  247. [`SECURITY.md`](./SECURITY.md).
  248. Copyright on `yt-dlp` downloads and music beds is on you. Brand marks from
  249. Simple Icons / Lucide are for editorial use; check trademark rules before ads.
  250. `review.py` binds localhost by default. Only pass `--lan` on a network you
  251. trust. No auth on that server.
  252. ## Credit
  253. Composition and rendering by
  254. [HyperFrames](https://github.com/heygen-com/hyperframes) (version pinned in
  255. `scripts/compose.py`).
  256. Designed scenes are [Remotion](https://www.remotion.dev) compositions. That
  257. layer's shape — a spec approved before anything is built, one composition per
  258. scene, frame-driven motion, no background music — is adapted from
  259. [Creatorberry's flick](https://github.com/Creatorberry/flick) (MIT), which does
  260. the same job standalone and is worth using directly if a script is all you have.
  261. MIT. Vulnerabilities: [`SECURITY.md`](./SECURITY.md).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.