Skip to content
Back to search

video-editor-agent

Claude Code skill pack: edit short-form videos end to end — style-clone a reference reel, build a branded motion-graphics edit, ElevenLabs sound design, frame-level QA, timeline-comment review loop

Safety scan

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

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/mrpaidsocial/mrpaidsocial-video-editor-agent with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/mrpaidsocial/mrpaidsocial-video-editor-agent'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/krusemediallc--video-editor-agent/badge.svg)](/listings/krusemediallc--video-editor-agent)

HTML

<a href="/listings/krusemediallc--video-editor-agent"><img src="/api/listings/krusemediallc--video-editor-agent/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.

1. Clone this repo and open it in Claude Code. 2. Run `bash scripts/setup.sh --demo`. It installs local QA dependencies, creates missing config templates and builds a synthetic edit, catalog, QA evidence and versioned review page. Requires Python 3.10+, Node.js 20+, ffmpeg/ffprobe; no paid APIs or model downloads. **[SETUP.md](SETUP.md)** covers prerequisites, workflow-specific setup and safe skill linking. 3. Fill in your brand, defaults and **projects directory** in `MASTER_CONTEXT.md` (where the videos live — default `outputs/`). Configure keys only for workflows you use. 4. Drop your raw footage into the projects directory (or `footage/`) and say: **"edit this like `<reference reel>`"** — or just "edit this video". The `video-edit-pipeline` skill takes it from there.

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.

README.md · README · 177 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. # Video Editor Agent
  2. A Claude Code skill pack that edits short-form videos **end to end**: raw footage in, a
  3. finished edit out — style cloned from a reference reel, sound-designed with ElevenLabs,
  4. QA'd frame by frame and in dB, delivered on a live review page with timeline comments,
  5. and revised round after round until sign-off.
  6. This is not a rendering library. It is a set of skills, local project tools and a QA engine that make a
  7. Claude Code session behave like a working video editor with a proven pipeline and a
  8. "verify pixels and dB, not intentions" culture. It gets better with every real edit: the
  9. process goes into the skills, the personal details stay in a gitignored context file.
  10. ## The pipeline
  11. ```
  12. footage + (optional) reference reel + brand/site + ratio
  13. │
  14. ▼
  15. [0] video-edit-pipeline picks the lane
  16. talking head + brand ─────────────→ [1]–[7] below
  17. creator's own reel look ──────────→ reel-recut
  18. multi-take screen+camera+mic ─────→ arcads-video-edit
  19. event recap: takes + B-roll ──────→ recap-video
  20. one long file of many hooks ──────→ hook-splitter (→ hook-variations)
  21. AI-actor footage that sounds fake → ai-audio-sound-design
  22. │
  23. ▼
  24. [1] reel-style-clone ──────────→ STYLE-GUIDE.md (only if a reference exists)
  25. │
  26. ▼
  27. [2] branded-ad-edit
  28. ingest → two crop candidates (zoom vs blur-pad — show both)
  29. whisper transcript → storyboard from the style guide
  30. hand-authored composition + karaoke captions + SFX markers
  31. B-roll: broll-capture (real screens) / openart-broll / arcads-broll (generated)
  32. │
  33. ▼
  34. [3] sound-design ──────────────→ ElevenLabs SFX kit + music bed, mixed by math
  35. │
  36. ▼
  37. [4] QA loop (video-qa + tools/video-qa)
  38. hyperframes check → ONE multi-timestamp snapshot → LOOK with vision
  39. → fix → render MP4 → the engine verifies the RENDERED file (frames + dB + seams)
  40. │
  41. ▼
  42. [5] video-review-canvas ───────→ live review URL (reply leads with the link)
  43. │
  44. ▼
  45. [6] revision rounds
  46. read notes → screenshot the exact frame per note → fix
  47. edl-tighten for pacing → new file per version → republish same slug
  48. per-note QA table with evidence from the rendered file
  49. one cut → many variants: hook-variations, then naming-convention
  50. │
  51. ▼
  52. [7] (optional) capcut-export ──→ CapCut draft for a human's final pass
  53. ```
  54. ## Quickstart
  55. 1. Clone this repo and open it in Claude Code.
  56. 2. Run `bash scripts/setup.sh --demo`. It installs local QA dependencies, creates missing
  57. config templates and builds a synthetic edit, catalog, QA evidence and versioned review
  58. page. Requires Python 3.10+, Node.js 20+, ffmpeg/ffprobe; no paid APIs or model downloads.
  59. **[SETUP.md](SETUP.md)** covers prerequisites, workflow-specific setup and safe skill linking.
  60. 3. Fill in your brand, defaults and **projects directory** in `MASTER_CONTEXT.md`
  61. (where the videos live — default `outputs/`). Configure keys only for workflows you use.
  62. 4. Drop your raw footage into the projects directory (or `footage/`) and say:
  63. **"edit this like `<reference reel>`"** — or just "edit this video". The
  64. `video-edit-pipeline` skill takes it from there.
  65. ## Project and review tools
  66. - **Review canvas v2:** version picker, synchronized comparison, open/resolved notes,
  67. replies and before/after frame evidence. Shared reviews use append-only here.now records;
  68. an explicitly local mode supports demos without hosting. Export the complete revision
  69. ledger as JSON and import it into project state.
  70. - **Storyboard coverage:** `qa:storyboard` checks planned IDs and timed HTML intervals,
  71. then extracts contact sheets from the actual MP4. Static coverage and visual verification
  72. are reported separately; see [the QA engine](tools/video-qa/README.md).
  73. - **Resumable projects:** `python3 tools/editor/editor.py project init|status|resume`
  74. records source identities, stage artifacts, renders, approvals and outstanding notes.
  75. Changed or missing files invalidate the relevant progress. Explicit configured commands
  76. can resume one stage at a time.
  77. - **Reusable footage catalog:** `python3 tools/editor/editor.py catalog ingest|search`
  78. indexes media metadata, thumbnails, local transcripts, selected intervals and past uses.
  79. Stable content IDs prevent duplicate indexing. See [project tools](tools/editor/README.md)
  80. for full commands and transcript options.
  81. - **Offline regression suite:** `bash scripts/test.sh --integration` covers production
  82. failures, project/catalog behavior, setup safety, the review model and a real media demo.
  83. GitHub Actions runs the same suite on Linux. Paid/live transcription is opt-in.
  84. QA reports include layer coverage; unavailable checks are not labeled a clean pass.
  85. Review instructions and sampling configuration participate in cache identity. The recut
  86. renderer uses bounded lossless batches and one final AAC encode for long cut timelines.
  87. Serve any local review with `python3 scripts/serve-review.py --directory <review-folder>`;
  88. it supports the byte-range requests video players need for frame seeking.
  89. ## Using it from your own working repo
  90. Most people keep their videos, queues and brand tooling in a repo of their own. Keep that,
  91. and point it at this pack instead of copying skills into it:
  92. - Link the skills with `bash "<pack>/scripts/setup.sh" --no-install --link-skills "<working-repo>"`.
  93. Setup computes relative links for the actual directory layout and preserves existing
  94. paths. A session started in your repo loads the skills; their files stay here.
  95. - Set the projects directory in `MASTER_CONTEXT.md` (and `VIDEO_PROJECTS_DIR` in `.env`)
  96. to your media folder. Media never moves into this repo.
  97. - Alias the QA engine in your `package.json`:
  98. `"qa:video": "npm --prefix \"<path to this repo>/tools/video-qa\" run qa:video --"`.
  99. Relative paths resolve from where you run it; your `.env` is read first.
  100. - Improvements from every real edit go back into the skills here — generic process only.
  101. Your client names, fees, reviewer preferences and machine paths belong in
  102. `MASTER_CONTEXT.md`, which never leaves your disk.
  103. ## Prerequisites
  104. | Dependency | Why | Install / notes |
  105. |---|---|---|
  106. | Node.js >= 20 | HyperFrames, the QA engine, the scripts | nodejs.org |
  107. | ffmpeg / ffprobe | every probe, extract, crop, mux | `brew install ffmpeg` (or your package manager) — no libass/drawtext needed |
  108. | HyperFrames | composition + rendering engine | `npx hyperframes`; then `npx hyperframes skills update talking-head-recut` (pulls fonts + gsap) |
  109. | whisper (bundled route) | word-level transcription | `npx hyperframes transcribe` manages whisper.cpp models; `whisper-cli` + a ggml model unlocks the VAD-driven cut planners |
  110. | python3 | helper scripts | PIL (`pip install pillow`) for overlays; numpy + scipy for `ai-audio-sound-design` |
  111. | QA engine | `tools/video-qa` | `npm --prefix tools/video-qa install` (tsx, zod, dotenv) |
  112. | ElevenLabs API key (sound generation only) | SFX + music + ambience generation | `ELEVENLABS_API_KEY` in `.env`; local/demo/QA workflows do not require it |
  113. | here-now skill + credentials | canvas review delivery | agent docs at https://here.now/docs (fetch with `User-Agent: claude`); credentials live in `~/.herenow/credentials` |
  114. | `GEMINI_API_KEY` (optional) | video-qa's watch+listen layer (L3) | skip if unset; the other layers still run |
  115. | `OPENAI_API_KEY` (optional) | cloud whisper fallback (QA engine, hook-splitter, arcads-video-edit) | local whisper.cpp is the default |
  116. | Puppeteer (optional) | scripted website screenshots/B-roll (`broll-capture`) | `npm i puppeteer` in the repo; or use a connected browser MCP instead |
  117. | Screen Studio (optional) | high-fidelity real-browser B-roll (`broll-capture` Lane C) | any screen recorder works; Screen Studio + its CLI is the polished path |
  118. | OpenArt MCP (optional) | AI-generated B-roll / talking heads / overlays (`openart-broll`) | connect the OpenArt MCP in your client; verify with `openart_account_get` — no API key |
  119. | Arcads (optional) | generated B-roll (`arcads-broll`) and the `arcads-video-edit` clip lane | `ARCADS_API_KEY` / `ARCADS_BASIC_AUTH` in `.env` + clone [arcads-claude-code](https://github.com/krusemediallc/arcads-claude-code) alongside |
  120. | Swift toolchain (optional) | `hook-variations`' AVFoundation probe (`avtest`) | Xcode command-line tools; built on first use |
  121. | pyJianYingDraft in a venv (optional) | CapCut draft export | only needed for the capcut-export handoff |
  122. External dependencies are documented, not vendored — nothing in this repo ships a copy of
  123. HyperFrames, whisper models, or ffmpeg.
  124. ## Skills catalog
  125. | Skill | What it does |
  126. |---|---|
  127. | `video-edit-pipeline` | **Master orchestrator.** Picks the lane and routes any edit request through the full pipeline. Start here. |
  128. | `branded-ad-edit` | Raw talking head → finished branded motion-graphics ad: framing grammar, card-per-line, karaoke captions, ~40 SFX + bed, QA → render → verify. |
  129. | `reel-recut` | The creator's own short-form look from ONE JSON spec: title banner, karaoke captions, callout boxes, silence-cut pacing; raw-cut mode for footage a client's editor finishes. |
  130. | `reel-style-clone` | Reverse-engineer a reference reel frame by frame into a STYLE-GUIDE.md + build directives. |
  131. | `arcads-video-edit` | Multi-take screen + camera + mic recordings → an EDL-driven base cut the reviewer locks, then a HyperFrames motion-graphics pass with takeovers, the base video as a character, captions, SFX and music. |
  132. | [Recap Video](.claude/skills/recap-video/SKILL.md) | Event/conference/travel recap: choose the best take per line, remove pauses and restarts, match supplied B-roll to narration, mix split/full layouts, verify and revise. Includes a portable base-cut assembler. |
  133. | `hook-splitter` | One long recording of many hooks/takes → one tightened standalone video per hook, QA'd by re-transcribing the renders, delivered on a gallery canvas with a comment box per video. |
  134. | `hook-variations` | One approved body × N hooks → N standalone variants, joined losslessly, loudness-matched, verified with AVFoundation (not just ffmpeg). |
  135. | `edits-hook-overlay` | Instagram's native text hook (Edits font, boxes, anti-aliasing) typed once on a USB iPhone over a black/white plate, lifted as an exact transparent PNG and burned onto any number of videos; HDR phone footage stays untouched. |
  136. | `naming-convention` | Filenames that carry every axis that varies; verify a subject label before baking it into 40 files. |
  137. | `edl-tighten` | Surgical silence/pacing cuts with a full timeline remap (captions/cards/SFX stay synced). |
  138. | `sound-design` | ElevenLabs SFX/music generation, audit pass, style-matching a reference track, mixing math. |
  139. | `ai-audio-sound-design` | Rebuild the audio of AI-actor footage: location ambience beds, room-matched reverb, outdoor distance, censor bleeps, watermark-whine removal, social loudness master. |
  140. | `video-qa` | The 4-layer QA procedure — and `tools/video-qa`, the engine that runs it on any rendered MP4. |
  141. | `video-review-canvas` | here.now review page with frame-accurate scrubber + timeline comments; reads notes back per version. |
  142. | `broll-capture` | Website screenshots + B-roll: Puppeteer/Playwright-MCP full-page shots, scripted-scroll recordings, and a Screen Studio real-browser lane for automation-blocked sites. |
  143. | `openart-broll` | GENERATED B-roll, identity-referenced talking heads, and screen-blend overlays via the **OpenArt MCP** — the pipeline's primary generation lane. |
  144. | `arcads-broll` | Alternative generation backend via the Arcads external API (REST) — companion pack: [arcads-claude-code](https://github.com/krusemediallc/arcads-claude-code). |
  145. | `capcut-export` | Layered export into a CapCut draft via pyJianYingDraft. |
  146. ## Limitations (honest)
  147. - **CapCut export is work-in-progress.** The layered export works but the schema patch is
  148. still being hardened — always verify the draft opens in CapCut before relying on it.
  149. - **Shared canvas delivery requires here.now.** Local mode runs without hosting and keeps
  150. feedback in that browser; export its JSON to transfer notes. Shared mode persists feedback
  151. through the existing public review link. Local mode is not a multi-device review store.
  152. - Transcription quality tracks your whisper model choice; tiny models miss words that then
  153. miss captions.
  154. - The main build assumes single-subject talking-head source footage; multi-shot sources go
  155. through `arcads-video-edit` (screen + camera takes), `recap-video` (spoken takes and
  156. event B-roll), or `hook-splitter` (one long composite) first.
  157. - Default tests are offline. The optional live transcription integration uses macOS `say`
  158. and an installed transcriber; deterministic seam tests run without it.

Ratings and comments

Sign in to rate this skill.

Sign in

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