heditor

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid plan, and nothing else to install first. Installing is only copying files. Using the skill afterwards is a separate, much heavier step with its own prerequisites, none of which you may install or run during this install: Linux or macOS (Windows via WSL2), Python 3.10-3.13, Node.js 18 or newer, ffmpeg/ffprobe, git and curl, about 1.6 GB of free disk for the Python environment, models and audio library (plus about 99 MB for the optional depth model), and 8 GB of RAM recommended. The one-time machine setup the skill itself documents (scripts/setup.sh, at use time only, never during this install) installs system packages with sudo, creates a Python venv in ~/.heditor/venv, installs packages from PyPI, fetches HyperFrames and a Chrome build via npx, and downloads fonts, models, music and sound effects (every download is disclosed in point 2). I also provide the source video myself as a local file, and I fill in ~/.heditor/brand.json myself (brand name, @handle, colours, logo path, language) when I first use it; never invent a handle for me. The skill itself is free software under the MIT licence; the music and sound effects it downloads at setup are under the Mixkit Free License, and the Depth Anything V2 Small model is the Apache-2.0 one (the Base/Large variants are non-commercial and are not the ones used). Install "heditor" for me. It edits a talking-head, class, live or podcast recording into a vertical Reel/Short/TikTok (1080x1920) or a long-form video (16:9, 16:9 4K, 1:1, 4:5 or 9:16) with pauses removed, word-synced captions, motion graphics, sound design and music: it probes the source locally, transcribes it locally with faster-whisper (word timestamps), picks a self-contained cut at clean word boundaries without inventing speech, designs an edit for that specific video in one of 37 named styles (for example cinema, podcast, tribunal, keynote, nativo — a clean cuts-and-captions-only edit) or in creative mode (an original design researched for that video, or a licence-safe homage to a reference such as GTA 6), builds an HTML + GSAP composition rendered to MP4 locally by HyperFrames, mixes voice-first at -14 LUFS, and runs local QA gates (lint, previews, tests, loudness, ASR of the final mix, contact sheet) before delivering the local MP4 path. Long-form mode adds chunked resumable transcription, a reviewable EDL proposal, chapters, layouts, SRT/VTT export and a segmented resumable render. Layered depthkit effects (text behind the speaker, depth maps, hand tracking) run locally with MediaPipe/Depth Anything models. Repository: https://github.com/pycodebr/heditor (the skill is the whole skills/heditor/ folder at the pinned commit: SKILL.md plus its references/, scripts/, templates/ and assets/ folders, 90 files) Pinned version: install the code at commit c7e4d53f0af404315a5fe6b5c119d8b2175be1e6 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/pycodebr/heditor into a temporary folder, check out commit c7e4d53f0af404315a5fe6b5c119d8b2175be1e6, and copy the skills/heditor/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 7771157469b3e983e000165ae6c52a8d7df3900190742205b0fc41dd0b98261f - assets/catalog.json: SHA-256 e7e87b4e5be7eee7021c3950b2341eb8e591c2e2ad0e5160ce226cb7fc901028 - references/adding-styles.md: SHA-256 4ff7c13e8a6e59b2d0dcccaec6ea932a9a3c640b2abe00da3d4c9e0941f32c6e - references/captions-and-layout.md: SHA-256 d08046e5c37a0c1a4edcff436bff118400661d9942159a3e14a48bc095b33d5f - references/creative-mode.md: SHA-256 27df0de312a1d4027002a0bda692b612a8d9a1fa269a1d592920c5a89726caed - references/depth-and-3d.md: SHA-256 8aadf1869c15940efb04704545d636e95ba8ecfeb200d498b6d811a6312bcbcf - references/kit-api.md: SHA-256 e4bc213f828e4055bca5e34eed0b0a57575a5b5cb4a1a90d116d4f848a4b557e - references/licenses.md: SHA-256 45d0e50d9b97000af576df39c5b2026fe88e970026b2c7414577caf78a1c4b32 - references/long-form.md: SHA-256 aa523154154578957477346049f3422f7c22b0692115a33a3c1c87cbaca8ea44 - references/quality-gates.md: SHA-256 74001a7d38b887a955bb2db1ef258c33a9bd7395bb46b083d164282cdc54d182 - references/sound-design.md: SHA-256 2f14dea71b9d26bf071af7a686968ba4fdd0c8fc6a2a0ed907fdebda8cf4714a - references/styles/INDEX.md: SHA-256 5f354878630d797c6c673b4c68004743c22234c5750eac253572060ec6538387 - references/styles/aduana-de-dados.md: SHA-256 6c19fbb02a7fcded7c9605e5c78762b2fbbf9061de6d90e892abadc8591918c3 - references/styles/aeroporto.md: SHA-256 91c96d124df74211ae40d87f7aa7b6ae84d26caf07cd25cc53d7eeb2b8863089 - references/styles/album.md: SHA-256 0d3dea1df0517cce8cdda18dffc0dcb47493d4bab8c3f64c92be973c33681fe6 - references/styles/arquivo-cofre.md: SHA-256 7806b8059aa2b0b1ea01138ab43cd7bdebe2df8bb42cb1032bf0ff27c551a4c4 - references/styles/auditorio.md: SHA-256 87fe5b5b5a10bd2f98b54f39f6df138ece3b5098eb05aa7e36c412bf49d15e42 - references/styles/bancada.md: SHA-256 e3f5cde7f60063c648a08c2348151677cd7df4614d1be83e326bc638bfa314f9 - references/styles/bilheteria.md: SHA-256 7f1b4061283f3f8214458cd57205adbe8cb14be921d20042416b73129242cfb0 - references/styles/camadas.md: SHA-256 f934b4731d3ac99160e0613913b78f68798eb539552d4ae4084552fd628a6c47 - references/styles/carrinho.md: SHA-256 21bb059f47bcb11446ca866a6f4a9369ecd22e7d89256109a98afaee1b51f253 - references/styles/cassino-de-precisao.md: SHA-256 e6a81992c407fb48615cb232b1ede46e47458aaaf0c453303324e97f5b2c210f - references/styles/chat.md: SHA-256 c84fd289aaa28c9d48af81653631ee4828c99502dc3261c504f7b0b5b67fb725 - references/styles/cinema.md: SHA-256 6530cba68c103e512ad0d69f8141364b316b88c20b9f63083a1f0d6659744743 - references/styles/documentario-narrado.md: SHA-256 08e40725e8d51221f0ce17f83515973f8a07bfecd18d9dc0f4e3ffc45ea1743f - references/styles/documentario.md: SHA-256 7bcd0c21dbfccefb8b0951e862e8c643e151d84ce99a7516629777a965e122fb - references/styles/duelo.md: SHA-256 b478c8f97c3437de3ed92eeed5e7f86bcb4a533d13703d90011191de59e38f9c - references/styles/entrega.md: SHA-256 fe7ce688e67acad1c743a6153b165d4fbb504afde0951f7ca1d7033966aaba1c - references/styles/game.md: SHA-256 7b490e7a387423810df40064c952ad8444345bae5986996860be395f6c86b99b - references/styles/ide.md: SHA-256 5f2a58eb9180472b8ada89583ec62e4f43241b3248ec598b595be12190b36284 - references/styles/impacto.md: SHA-256 92f1b10af19c3412470ca0c99772e227251271307246b9faf27078faaabd3760 - references/styles/keynote.md: SHA-256 f61814fbe977a5b2e789681729abcc1119dc53e2664bb74308daa7d461a8fa1f - references/styles/laudo-de-sinistro.md: SHA-256 616b625280810da3b2fb47b6a3203686b5b214d1f73a6e0d200c8c3858b49906 - references/styles/lousa-de-aula.md: SHA-256 6a5c421af6d7359fd78be3dc316a9a64a1fee3961346c505796d0e031873d8ae - references/styles/manchete.md: SHA-256 95bfabd6747b546a0fa7a454a2116236830248379de42341e977279c6d8dcc79 - references/styles/manual-de-montagem.md: SHA-256 36aa9656891dce837eb48374bd7b0d8126a55c05273e9c6f08245fb15994761c - references/styles/mesa-de-risco.md: SHA-256 edd00acde65214d49263c973ac22a25a0a87793ab2a52f7b4c3b83cb036ac7ff - references/styles/missao.md: SHA-256 5862b86bb9bed22443af3c8a5f2e5c1dc69817981353cb8a503372ce8eda5dfa - references/styles/nativo.md: SHA-256 48e0218ef3796175d7fc144c403b49ecbae012c5bb9194670c65c4838653a9ac - references/styles/podcast.md: SHA-256 13a219c7204cce14c1a2a95b77734b2cc5be297ca4e511e3d26135cf8d7f15ea - references/styles/porta-de-aco.md: SHA-256 bce629dc61c98887ec243418d251ff4e21657fefeecb6bd0724513ee624d0c10 - references/styles/predio-24h.md: SHA-256 68edc6429ac60f8158796ace462bcda4d76ddbffec14597d1bab9a50f87b5990 - references/styles/raio-x.md: SHA-256 22b4981e6971670de359345b8451d1daa195b01aa78d7cc0322ae46678676b8b - references/styles/revista.md: SHA-256 ad32fcb44278175f66112847aad2a34544a5c93fc83346752ea7ab2ce153ec5a - references/styles/ruptura.md: SHA-256 c04695a9fa82aec392ca269dc3e6b1bebdae6161d7172719af9bc92d38ae2682 - references/styles/stories.md: SHA-256 bcb578316a0856d5c9646f85dc221f203f6551a6195c04c4aaa53125adec1552 - references/styles/thriller.md: SHA-256 37e9968c0b3f55583b1ac9a625dc71f55b01af5d0693445f324025294d2c5223 - references/styles/tribunal.md: SHA-256 15f0866e69befc64051b850fb09cbc6f47b5a1e172c93ebb77617de274f301bf - references/styles/turno-da-madrugada.md: SHA-256 7bbd3588e95ca5f4b5fe30d56822c64b207c746eaca09a88c6f7ca61d7ea30b1 - references/technique-catalogue.md: SHA-256 812cdf96bce2a1e7c6dbce064f02a4d04b0f84eaec8dbba227716f5e9a7ac936 - references/troubleshooting.md: SHA-256 f6b107c7a7814de2256ce53afbb2efdf70f3606ec957889acaa9be29136fb9bc - references/workflow.md: SHA-256 9b93bb5df2949ce1467800ea70967ae600ebb6186df9d288ff2ff42dbee367fc - scripts/doctor.py: SHA-256 8a4080a491dee9d173cc4ccd048eed8ad993a51512b41f3c503ce5bae3722a4f - scripts/fetch_assets.py: SHA-256 8c26494ad2f24a91f0a63ed637bb185dd9d783f7e0f0b76fcf32fe81a115efd4 - scripts/heditor: SHA-256 7fe747971dab53d698704c2d1dd0caadd9bda53595e9fd039d98782c0bd27fcd - scripts/reelkit/asr_check.py: SHA-256 91e35b979d4d7b8fbed342c7189ee8fe4cb9d3ac8c7e73de720027f5f1ae336f - scripts/reelkit/asr_window.py: SHA-256 c915df20e3ba29cdad0c1ef61a3714c967078cbc4ce3520df5f7ffac7e948ac0 - scripts/reelkit/camera_sync_check.py: SHA-256 5f54f2c34d987b5a1ead0c54ec3e8b3e58471b7ea43ccc6db6d515a7d020fd8a - scripts/reelkit/depthkit.py: SHA-256 d25d1834c0772226a7cdfc05c7195a41c2047fcfc4dcfec47a5cb924ec516963 - scripts/reelkit/finalize.py: SHA-256 3cfa6e7612b401a5167a13bd0ddfc2a60791bf4812382ef6bd820f1533440876 - scripts/reelkit/formats.py: SHA-256 f6b84e9b8ca6764baf93cceaf43aaba3ef0dd8c741d9f23cd29cccf078dfbc82 - scripts/reelkit/gaps.py: SHA-256 91612ccf3310093c29468072ba6d088ca07fc4163ba07d08515280a4c638fcfd - scripts/reelkit/layers.py: SHA-256 378207b25bf997ba7fe8a4549a9e967e85afa60caba0ce9c51f47b519c88c5ae - scripts/reelkit/long_cli.py: SHA-256 7af1bddab5cf3f400c0d0ebccd3c19dae23c494e4e5dcab891ec8b6c813457d1 - scripts/reelkit/longform.py: SHA-256 44e77dc1b48cbc658ffb65c7f8fb6a404e3ec4f25a4ca5475a115cdbd9a50754 - scripts/reelkit/longkit.py: SHA-256 f4e4469fd99bfecfdc2b827025cc82a6d6f452915992abf398e7e515ef8b3b83 - scripts/reelkit/longmix.py: SHA-256 b185d111444f229b6132a352c78cf0cc167fb4c83576966c6ffc14e7de85960d - scripts/reelkit/longqa.py: SHA-256 2fde99cb7c3117e995ad3e87de1e31d488622c28153d0238da325233d5266b58 - scripts/reelkit/prepare_cut.py: SHA-256 3085685d28eb00a9a3bc5a2d700e083ab6a98a0c1ebdefdc3b0b1e2d6e967e91 - scripts/reelkit/preview.py: SHA-256 fca1a0c3cc1bc7996e436861168ae2573c202114b975cc94d08599d1d2034bad - scripts/reelkit/produce.py: SHA-256 0f38173f658569ac02b6d39d7d23d04ae4969793fb0e742e1c443d88a875cca3 - scripts/reelkit/reelkit.py: SHA-256 6abcf7e095d00aec4fb03a9a180298d626b237142ce467a29103bee7a7beeb56 - scripts/reelkit/stylepacks.py: SHA-256 a94754882ecfa7679cf2b1c93529dcdbdea3c8f4bfa844cc504bafa41a241625 - scripts/reelkit/test_cut.py: SHA-256 44c57bec2e051f1fcded380e4d0e7c2df289a61bbde617511e31c31f9dd57eff - scripts/reelkit/test_depthkit.py: SHA-256 349446cf78bc90f6900ddb8410688897cb408798dde06305894b10d6b9238d05 - scripts/reelkit/test_formats.py: SHA-256 445199938219155f62f261d8a7e9bf6dcfb43962733ac17db2f8eefada1ab5c0 - scripts/reelkit/test_longform.py: SHA-256 f1872a8142c30bacd13eaae04628efc1e1e67020ca65fa65f29261ab922a5022 - scripts/reelkit/transcribe_cut.py: SHA-256 ed5923d7951ce23835e3777070abe59a9e189ff9710cc1b550d312f4b95c5119 - scripts/requirements.txt: SHA-256 31c76ad22a13e39951fea1972a53288c5c7518e4f888e62084cec567b751973a - scripts/setup.sh: SHA-256 9714d4193dd741acf78ff764ace8ba0ca2cebc0373bbb7c0d576d7595076ae9b - templates/DESIGN.md: SHA-256 be04417f947477b19acddaa30aca2bf8d185c4866fc4c21d24eaa7b2b4dae160 - templates/brand.example.json: SHA-256 436da7a932a6463e689921d7154668808b9550992370b897f22cdae67efa266f - templates/build.py: SHA-256 1f95b493a226d5eb6075471c03ccc6694f077f5b091d4519b9b690532a8d0605 - templates/composition-package.json: SHA-256 0c9fcb2742abc2d84bd95687e915a7c074da797a0b0cf08b8b6211d35a6c1a2f - templates/long/DESIGN.md: SHA-256 11a171ecc7a505095c31fc106840190a6ba4b082400375f8cea8cbb3f714b46f - templates/long/build.py: SHA-256 933f58e25de1896a6f4fbb94d2e0b84f122591ede4c36812628d3a05c4aa208b - templates/long/chapters.json: SHA-256 094ebeffb949551b6bbd0e79dc73071491af1ba4d2809232f59894980f3b08dd - templates/long/long.json: SHA-256 27f1ddebeea3920adc44f346e21700caed916caeb117be49df26e2d86f8b5418 - templates/prepare.py: SHA-256 384591f8198b4585b5fdda4d4a353efa2bc23ea5b2522ad59dcf7e40ff8a9e8e - templates/styles/nativo/build.py: SHA-256 3b9ccd7b483e445ff721e08ca7e062472ef08f7bfabf96a66642f5f1de229e4b If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full and read the scripts in scripts/ (setup.sh, doctor.py, fetch_assets.py, the heditor CLI and the reelkit/ Python kit), and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 90 files only: SKILL.md (17,361 bytes), references/ (the workflow, long-form, creative-mode, kit, depth, sound, captions, quality-gate, licence and troubleshooting guides plus one file per named style), scripts/ (setup.sh, doctor.py, fetch_assets.py, the heditor shell CLI, requirements.txt and the reelkit/ Python kit), templates/ (prepare.py, build.py, DESIGN.md, brand.example.json, composition-package.json and the long/ and styles/nativo/ templates) and assets/catalog.json (the Mixkit music/SFX catalogue: 18 music tracks, 85 sound effects, URLs only — the audio files themselves are not in the repository). The folder is self-contained for copying: it needs no file from elsewhere in the repository — not the root README, not the root install.sh and not the other root files. But the copied folder does nothing on its own: at use time it relies on a machine state that lives outside the folder, in ~/.heditor/ (venv, env.sh, brand.json and the downloaded assets and models), plus system tools (ffmpeg, Node.js, Python) and the HyperFrames npm package — all of that is created by the use-time setup disclosed below, not by this install. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. Do NOT run anything from the folder during this install: no scripts/setup.sh, no doctor.py, no fetch_assets.py, no heditor command, no pip install, no npm/npx install, no asset or model download and no rendering. Never install this skill through the upstream curl-pipe route (curl -fsSL https://raw.githubusercontent.com/pycodebr/heditor/main/install.sh | bash): that installer is outside the pinned skill folder, follows the moving main branch instead of the pinned commit, links the skill into every detected agent and immediately runs the heavy setup — none of that is this install. - What leaves or reaches this machine at use time, disclosed here before it ever happens — using the skill is a separate step that only starts when I ask you to edit a video, and the setup below runs only when I ask for that first edit (or when I run it myself): (a) One-time machine setup, scripts/setup.sh: it installs missing system tools (curl, git, ffmpeg, Node.js, Python) through my package manager (apt-get, brew, dnf or pacman), using sudo when I am not root; on apt systems where Node.js is missing or too old it pipes the NodeSource setup script straight into sudo bash (curl -fsSL https://deb.nodesource.com/setup_22.x | sudo bash) before installing nodejs — flag that to me before running setup so I can decide; it creates ~/.heditor/venv and pip-installs scripts/requirements.txt from PyPI (numpy, pillow, faster-whisper, mediapipe, opencv-contrib-python, onnxruntime, playwright, pytest); it runs npx --yes [email protected] (npm registry download) to fetch HyperFrames and ensure its Chrome build; and it runs scripts/fetch_assets.py, which only downloads (HTTP GET, with a User-Agent of "heditor-setup/1.0 (+https://github.com/pycodebr/heditor)" and no other data from me): fonts and their OFL licences from raw.githubusercontent.com/google/fonts, GSAP 3.14.2 and three.js r181 from cdn.jsdelivr.net, the Lucide and three.js licences from raw.githubusercontent.com, MediaPipe face/person/hand models from storage.googleapis.com, the 18 music tracks and 85 sound effects listed in assets/catalog.json from assets.mixkit.co, faster-whisper models from Hugging Face (downloaded through the faster-whisper library), and — only with --depth — the Depth Anything V2 Small model (about 99 MB) from huggingface.co. Setup also writes ~/.heditor/env.sh, copies templates/brand.example.json to ~/.heditor/brand.json if none exists, symlinks the heditor command into ~/.local/bin and the kit into ~/.heditor/kit, and finishes by running doctor.py (which itself calls npx --yes [email protected] --version). (b) First-use downloads after setup: depthkit downloads any missing MediaPipe segmenter/hand model from storage.googleapis.com and the depth model from huggingface.co on first use; reelkit downloads a missing Lucide icon SVG from raw.githubusercontent.com with curl when a composition asks for it; faster-whisper downloads its ASR model from Hugging Face on first use if setup did not pre-download it; some styles also tell you to download extra Google Fonts families (OFL) from github.com/google/fonts into the work folder. (c) Creative mode: SKILL.md and references/creative-mode.md tell you to research the video's subject or reference (palette, typography, UI, sound) with your own web search/extract tools, and to look at reference images if you can see images — the search queries go to whatever search service you already use, under its own terms; that research is about the reference (for example a game or a broadcast style), never about me, and no part of my video is uploaded for it. (d) Everything else at use time is local: probing, transcription (faster-whisper runs on this machine — my video and its audio never leave the machine for ASR), cutting, mixing, HyperFrames rendering, Playwright/Chrome previews and all QA run locally; the finished MP4 is delivered as a local path and SKILL.md says never to publish to any social network unless I explicitly ask. There are no API keys anywhere in this skill: the only environment variables the code reads are HEDITOR_* configuration (home, kit and skill paths, language, ASR model and thread counts, Chrome and HyperFrames commands, locks and timeouts) and doctor.py reads only the HEDITOR_CHROME path; brand.json (my brand name, @handle, logo path, colours and words) is a local file read locally and never sent anywhere. No code in the folder reads .env files, SSH or cloud credentials, browser stores or keychains, and no code uploads, posts or sends my files, transcripts or videos to any server — every network call in the folder is a download of a named font, library, model, icon or catalogue audio file from the hosts listed above. - Do not create ~/.heditor, do not set or ask for any key, and do not edit any video during this install: machine setup and editing are my decision, later, when I actually ask for an edit. 3. When I say "yes, install", install it as a skill: copy the whole heditor folder into the agent's skills directory, in a folder named "heditor", keeping all 90 files (SKILL.md at its root and the references/, scripts/, templates/ and assets/ subfolders intact). A manual clone followed by copying the skills/heditor folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as every installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: heditor), all 90 files in place with matching fingerprints. Do not run setup.sh, doctor.py, fetch_assets.py or any heditor command to verify. 5. Report what was installed, where, and what I still need to do myself (for example: when I first want an edit, let the use-time setup run — or run bash <skill_dir>/scripts/setup.sh myself — knowing it installs system packages with sudo, pipes the NodeSource script into sudo bash on apt systems missing a current Node.js, pip-installs the requirements from PyPI, fetches HyperFrames and Chrome via npx and downloads the fonts, models, Mixkit music/SFX and Whisper models disclosed in point 2, about 1.6 GB in total; then edit ~/.heditor/brand.json with my own brand name, @handle, colours and logo path; provide a source video as a local file for each edit; using the skill is a separate step that only starts when I ask for an edit, and the use-time downloads and the creative-mode web research disclosed above happen only then). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.