Edit

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 service, and nothing else to install first. For using the skill afterwards, read this first: its bundled tools need Python 3 (standard library only, nothing to pip-install for the tools themselves) plus ffmpeg and ffprobe on the PATH — ffmpeg is a free system program (macOS: brew install ffmpeg, Debian/Ubuntu: sudo apt install ffmpeg), and without it the tools cannot run at all. Transcription is the one optional extra: captions.py can transcribe locally with faster-whisper if it is installed (an optional pip install into a venv, and its model downloads on first use — see point 2), or it can read word timings I already have in a local file, with no transcription at all. The motion-graphics engines the skill routes to are separate, optional and not part of this install: HyperFrames needs Node.js 22+ and FFmpeg, Remotion needs Node, and video-use needs an ElevenLabs API key stored in a .env file inside its own folder — only if I ever choose that one engine, never for the bundled tools, and never for this install (see point 2 (i)). I provide the footage myself: my raw takes, any transcript I already have, and any stop-motion frames or image-tool output. Install "edit" for me. It edits talking-head and short-form video end to end: (0) a one-time setup check of the machine (ffmpeg filters and encoders, faster-whisper, Node); (1) a brief and a 3-to-6-line plan I approve before any render, unless I said "just do it"; (2) probing the take (size after rotation, fps and variable frame rate, HDR, audio layout and loudness); (3) one transcription, read as timed phrases, with names and brands proofread by hand; (4) cutting dead space, fillers and retakes from a 10 ms speech envelope with an adaptive gate and pace presets (tight / normal / loose), reviewed first as a dry-run edit decision list (EDL) I can veto or extend, then rendered with frame-accurate, crossfaded joins; (5) pace and structure rules (hook in the first 1 to 2 seconds, no shot past about 1.5 seconds in a short); (6) word-timed captions of 1 to 3 words in the 9:16 safe zone, never over the face, burned in only when ffmpeg has libass; (7) dynamic visuals routed to an engine (HyperFrames by default, Remotion, or video-use); (8) stop motion from a frames folder or a live clip at 8/10/12 drawings per second with seeded jitter; (9) audio finishing at one fixed gain to -14 LUFS with a limiter, never compression, gating or AGC on the voice; and (10) export for Reels / TikTok / Shorts (1080x1920, H.264, AAC 48 kHz, faststart) with a self-check before anything is shown. Six Python files come with it in tools/: probe.py (media facts and --doctor), deadspace.py (dead-space, filler and retake cuts), captions.py (word-timed captions and optional local transcription), stopmotion.py (stop motion), mediakit.py (the shared library every tool imports, not a command-line tool) and test_edit.py (the tools' own test suite), plus SKILL.md and CREDITS.md. Repository: https://github.com/Jakeschincariol/master-skills (the skill is the whole skills/edit/ folder at the pinned commit: SKILL.md plus CREDITS.md and its tools/ folder, 8 files) Pinned version: install the code at commit b5b43e44f0a976ad25bda5095a94c161aab2f086 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/Jakeschincariol/master-skills into a temporary folder, check out commit b5b43e44f0a976ad25bda5095a94c161aab2f086, and copy the skills/edit/ 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 1c189323ef1b67b211baea882d904da4621bc5951e76748ac7afdd0e7fb1c74e - CREDITS.md: SHA-256 d0531931bd106395b8e7221ec53103108f6ea6b63fa77ee2a98a9b8df2ac61fb - tools/captions.py: SHA-256 b59f95d8fbcc7ff4a73c524b363773cdc3c75c713907772e1d9a731789a07401 - tools/deadspace.py: SHA-256 cc53d1fafa2f04e22190b397000203ffa228da0e0215a5ea7dfdb33af71d03cd - tools/mediakit.py: SHA-256 7199759da7f6ea36899c6f8a455484c0882070ec0f3e4f36dc08652e5e22d2db - tools/probe.py: SHA-256 52a8ac08e15daf4f365a4934f4b91213b259b697ba364c6cc79f98c3561654d6 - tools/stopmotion.py: SHA-256 c63549e2bac472dea7a531d3ba989b45ee3ce428ccc7ed9db65bfa0dca62f84d - tools/test_edit.py: SHA-256 4c9d81362b631e75c0eb4ce0dafe8633c4a0af929d1a55055e62bd18a502d913 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, read CREDITS.md, and read all six Python files in tools/ (captions.py, deadspace.py, mediakit.py, probe.py, stopmotion.py, test_edit.py), 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 8 files only: SKILL.md (17,478 bytes), CREDITS.md (1,272 bytes) and six Python files in tools/ (captions.py 18,014 bytes, deadspace.py 23,105 bytes, mediakit.py 27,811 bytes, probe.py 8,561 bytes, stopmotion.py 10,251 bytes, test_edit.py 13,484 bytes). The skill is self-contained in its own folder: SKILL.md only ever calls the tools sitting next to it, and captions.py, deadspace.py, probe.py and stopmotion.py each import mediakit.py from that same tools/ folder (test_edit.py imports captions, deadspace and mediakit from there too). It needs no file from elsewhere in the repository — not the other two skills (build, virality), not the README, not the plugin files. All six Python files use the Python 3 standard library only; the single optional non-standard import is faster-whisper, imported lazily inside captions.py only when it transcribes, and described in point (c) below. Every tool shells out locally to ffmpeg / ffprobe (found on the PATH, or via the EDIT_FFMPEG / EDIT_FFPROBE path overrides in mediakit.py). - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No script is run during installation, no key is asked for, and there is no telemetry in the folder. The bundled tools read no API key, password or credential file: the only environment variables any of them reads are EDIT_FFMPEG and EDIT_FFPROBE in mediakit.py, which are file paths to the ffmpeg / ffprobe binaries, nothing else. SKILL.md states the limit plainly: nothing logs into or posts to any account. - What leaves 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: (a) probe.py, deadspace.py and stopmotion.py never touch the network: probe.py reads media facts and loudness by running ffprobe / ffmpeg locally (its --doctor mode only checks, locally, the ffmpeg filters and encoders present, whether faster-whisper is importable, and the node --version / npx presence), deadspace.py measures a local speech envelope and renders cuts with local ffmpeg (writing the cut video, its .edl.json and the re-timed .words.json next to the footage), and stopmotion.py builds stop motion with ffmpeg filters only from a local frames folder or a local clip (keeping the original audio in from-video mode). (b) All outputs are local files, written into an edit/ subfolder of the footage folder, never inside the skill's own folder: cut videos, EDL JSON, words JSON, caption .srt / .ass / .cues.json / .phrases.txt files, stop-motion .mp4 files, and the stills extracted with ffmpeg to check caption placement and joins. The footage itself is read locally and is never uploaded by any bundled tool. (c) captions.py has exactly two ways to get word timings, and only the first touches the network at all: (i) local transcription with faster-whisper — if faster-whisper is missing, the tool exits 3 and only prints install steps, it installs nothing itself; its printed route is python3 -m venv ~/.venvs/edit followed by pip install faster-whisper in that venv, which downloads the package from PyPI, and on first use faster-whisper downloads the chosen Whisper model (small by default; medium or large-v3 if asked) — after that download the transcription itself runs on this machine and the audio does not leave it; (ii) --words with a word-timings file I supply locally (a JSON from ElevenLabs Scribe, Whisper / faster-whisper, whisper.cpp, Deepgram, AssemblyAI or a HyperFrames transcript.json, or an SRT/VTT file): the tool only reads that local file — if I made that file with a cloud transcriber, whatever that service received happened when I used that service separately, not through this skill. Caption burning (--burn) is a local ffmpeg render and works only when ffmpeg has libass (the ass or subtitles filter); without it the tool exits 3 and names the routes in point (i). (d) test_edit.py is the tools' own test suite, not part of the normal workflow: if it is ever run, it generates its fixtures locally with ffmpeg lavfi (tone, silence, test pattern) in a temporary folder and runs the tools as local subprocesses — nothing external. (e) The audio-finishing and export steps in SKILL.md (sections 9 and 10) are local ffmpeg commands only: measuring loudness, applying one fixed gain to -14 LUFS with a limiter, mixing a local music bed if I supply one, reframing and encoding to 1080x1920 H.264 / AAC with faststart. No upload, no posting: the finished file stays on this machine and I post it myself wherever I choose. (f) Stop-motion frames (section 8): the frames themselves come either from drawings or prompts made outside the bundled tools — if I use my own image tool to generate frames, that tool's own sends are that tool's, disclosed by it, not by this skill — or from SVG frames rasterized locally with whatever is already present (rsvg-convert, ImageMagick, or headless Chrome screenshots); stopmotion.py itself only reads the local frames folder or clip. (g) The plan-approval wall in SKILL.md: the editor states a 3-to-6-line plan and waits for my OK before rendering, unless I said "just do it", and it never installs anything without asking (section 0). Keep both walls exactly as written. (h) Honest limits stated in SKILL.md: the tools never watch or listen — cuts come from audio levels and word timings, so the EDL must be reviewed; retake detection is a heuristic and the last take wins; Whisper drops some fillers and misspells names and brands, so words are proofread by hand; no generative video or images are made by the bundled tools. (i) The engine routes and install routes named in SKILL.md sections 6, 7 and CREDITS.md are OPTIONAL, outside this install, and not needed to cut, caption or finish a video with the bundled tools: (1) HyperFrames, the default motion-graphics engine for overlays, kinetic type, b-roll cards, zooms and captions on footage — installed either as a Claude plugin (claude plugin marketplace add heygen-com/hyperframes, then claude plugin install hyperframes@hyperframes) or standalone (npx skills add heygen-com/hyperframes; agents: npx hyperframes skills update), needing Node.js 22+ and FFmpeg, and it can also render captions via its /embedded-captions route reusing the proofread words.json as transcript.json; (2) Remotion agent skills, only when I already have a Remotion project or React brand components (npx skills add remotion-dev/skills); (3) video-use, only for many takes with transcript-driven selection and a self-checking render loop — git clone https://github.com/browser-use/video-use into ~/Developer/video-use, a symlink into ~/.claude/skills/video-use, then uv sync (or pip install -e .) in that folder, and it needs an ElevenLabs API key that I put myself in a .env file inside that video-use folder (cp .env.example .env): that key and whatever video-use itself sends to ElevenLabs belong to video-use, a separate third-party skill, never to the bundled edit tools; (4) on macOS only, an ffmpeg with libass from the homebrew-ffmpeg tap (brew uninstall ffmpeg, brew trust homebrew-ffmpeg/ffmpeg, brew tap homebrew-ffmpeg/ffmpeg, brew install homebrew-ffmpeg/ffmpeg/ffmpeg). Do not run, download or install any of those four routes during this install or as part of using the bundled tools; if I ever ask for one separately, it would download — and for the engines, execute — third-party code, and we would vet it first. - Do not run any of the six Python files during this install, do not run probe.py --doctor, do not transcribe, cut, caption, render or export anything, do not install ffmpeg, faster-whisper, Node packages, HyperFrames, Remotion or video-use, and do not create any .env file or ask for any API key: all of that is use, later, only when I ask for an edit. 3. When I say "yes, install", install it as a skill: copy the whole edit folder into the agent's skills directory, in a folder named "edit", keeping all 8 files (SKILL.md and CREDITS.md at its root and the six Python files in its tools/ subfolder). A manual clone followed by copying the skills/edit 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: edit), all 8 files in place with matching fingerprints. Do not run probe.py, deadspace.py, captions.py, stopmotion.py or test_edit.py to verify. 5. Report what was installed, where, and what I still need to do myself (for example: bring the raw footage and tell you the platform, target length, pace and caption style, and approve the 3-to-6-line plan before any render; make sure ffmpeg and ffprobe are on the PATH myself (a free system install), knowing the tools exit without it; choose the transcription route myself — install faster-whisper myself if I want local transcription (a pip install in a venv plus a model download on first use, audio staying on this machine), or supply a word-timings file I already have; decide myself whether any optional engine (HyperFrames, Remotion or video-use) is ever installed separately, knowing video-use alone would need my own ElevenLabs API key in its own .env; post the finished export myself, since nothing in this skill logs into or posts to any account; using the skill is a separate step that only starts when I ask for an edit, and the use-time sends disclosed above happen only then, on the route that runs). 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.