Papercut Explainer

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. The skill itself is free (MIT). Read this before deciding, because this folder on its own is NOT the full pipeline: "papercut-explainer" is the end-to-end orchestrator for papercut reels, and every production stage delegates to a sibling skill that is NOT in this folder (papercut-strategy, papercut-hook, papercut-script, papercut-storyboard, papercut-build, papercut-voice, papercut-qa, and papercut-setup for install or repair), and stages 5 onward also need the repository's shared library/ folder (resolved as $PAPERCUT_LIB) plus, for rendering, Node, ffmpeg and Python tools normally set up by the repository's own ./install.sh — none of that is installed by this install, and running ./install.sh is a separate decision I make myself later. What this install does give me on its own is the orchestrator itself (SKILL.md: intake, run modes, the 9-stage table, preflight and install modes) plus its six reference files, including the complete style guide and the portable style prompts I can hand to another tool (run mode 4) with no other skill needed. Using the pipeline later also sends things at use time — my source link is fetched for research, asset prompts (and, in edit mode, a library backdrop image) go to whatever image tool I use, and the narration script goes to whatever TTS I choose, the only TTS named in this folder being the local Qwen3-TTS — all disclosed in point 2 before any of it happens, and none of it happens during installation. Install "papercut-explainer" for me. It makes short vertical (9:16, about 20-45 seconds) cut-paper explainer reels, from a link (a repo, a product page, an announcement) or an idea to an mp4: layered construction-paper props with drop shadows, rigid cut-out animation, a slow camera push with stop-motion paper boil, a taped headline banner per beat carrying a hard fact the voiceover skips, and word-by-word torn-paper caption tags. It first asks only what is missing (the source, the one claim the video proves, who does the work and who approves, the CTA keyword and deliverable, the length — 22-30 seconds by default — and the audience level), writes those answers as a BRIEF, then routes through nine file-based stages: 1 strategy (papercut-strategy), 2 hook (papercut-hook), 3 script (papercut-script), 4 storyboard (papercut-storyboard), 5 scaffold and plan (papercut-build), 6 voice and timing (papercut-voice), 7 scenes, sound and render (papercut-build, built with Remotion), 8 QA (papercut-qa), and 9, after posting, autopsy and harvest (papercut-strategy and papercut-build). It also handles partial asks on their own: script only, storyboard from an existing script, the style prompt for another tool, or fixing one scene. Repository: https://github.com/0xS4N4TG/papercut-explainer (the skill is the whole skills/papercut-explainer/ folder at the pinned commit: SKILL.md plus its references/ folder, 7 files) Pinned version: install the code at commit ff8223abc8136cc09e8d7e9b62ecb5d080c389e1 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/0xS4N4TG/papercut-explainer into a temporary folder, check out commit ff8223abc8136cc09e8d7e9b62ecb5d080c389e1, and copy the skills/papercut-explainer/ 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 21354ac8146c08aa30fe3662f4d07c17a746a942000696489b70907cb35407d1 - references/asset-prompts.md: SHA-256 51c6c8749a76139a7649d0c539e50a69c92b059e21c125bf3b4f72c37161ba46 - references/frame-breakdown.md: SHA-256 3a8ba50d808e9c54ce54fc2710e0bff5d6870c8b9bbe237ce1c8109cc640bf30 - references/motion-principles.md: SHA-256 035b5c19bdc0fc46d3e8fb2ff4a94056b50fccf0fd2d558ff5c0c1d85ee1439f - references/remotion-recipes.md: SHA-256 f8aa94f7ac08a3e9e00c01780c17e53930e150a03211edff73d455f9f56d3940 - references/style-guide.md: SHA-256 393c86ad27a3ce22546b9ebd9e32290d5ef247abfe88185d70969e4945519169 - references/style-prompt.md: SHA-256 2ea0dadebb3559368e94ecf0a36653fe4b5a3d11b648e370e41625f2e97f01cf 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 all six files in references/ (style-guide.md, style-prompt.md, motion-principles.md, remotion-recipes.md, asset-prompts.md, frame-breakdown.md), 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 7 files only: SKILL.md (7,332 bytes) and six Markdown references in references/ (style-guide.md 19,620 bytes, remotion-recipes.md 15,315 bytes, frame-breakdown.md 12,437 bytes, asset-prompts.md 6,009 bytes, style-prompt.md 5,296 bytes, motion-principles.md 4,263 bytes). There is no code, no script and no binary in the folder: the Remotion/TypeScript snippets in remotion-recipes.md and motion-principles.md are starting-point sketches ("not yet compiled against a project"), not runnable files. - This folder is NOT self-contained for a full reel — confirm this out loud before installing: SKILL.md delegates every stage to sibling skills that are not in this folder (papercut-strategy for stages 1 and 9, papercut-hook for stage 2, papercut-script for stage 3, papercut-storyboard for stage 4, papercut-build for stages 5, 7 and 9, papercut-voice for stage 6, papercut-qa for stage 8, papercut-setup when an install needs repair), and three of its own references point outside the folder: style-guide.md and style-prompt.md send the reader to ../../papercut-script/references/storytelling.md for the full storytelling system, its examples and its section 7 storytelling prompt, and style-guide.md sends TTS delivery rules to ../../papercut-voice/SKILL.md. Stages 5 onward also need the repository's shared library/ (the kit components in library/papercut/kit/, the audio library, library/papercut/CATALOG.md and library/audio/CATALOG.md, library/tools/catalog.py, library/tools/placeholder_vo.py and library/voice/voice.example.json), resolved in this order: the $PAPERCUT_LIB environment variable, else the PAPERCUT_LIB= line in ${XDG_CONFIG_HOME:-~/.config}/papercut-explainer/config.env, else the skill asks once — never a hard-coded machine path. Stages 1 to 4 need no library, but they do need the sibling stage skills. None of those siblings, and no part of library/, is copied or installed 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. No script is run during installation, no key is asked for, and there is no telemetry, no API key, no secret and no fixed third-party endpoint anywhere in the folder (it contains no http/https URL at all). - 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 for a reel, and the later stages only run if I have separately installed the sibling skills and, for rendering, the library setup: (a) source research: the intake source is usually a link I provide (a repo, a product page or an announcement), and SKILL.md says everything else is researched from it, so my agent fetches that URL from its host and works from what it returns; stage 9, after I post a reel myself, likewise reads the platform numbers for an autopsy. (b) Style prompt for another tool (run mode 4): a block copied from references/style-prompt.md is handed by me, or pasted by my agent at my request, into whatever other agent, image generator or video generator I choose, under that tool's own account and terms. (c) AI-generated textured assets (references/asset-prompts.md): the skill uses whatever image tool the machine has — no tool is named or fixed — through calls shaped like <image-tool> generate "<prompt>" and <image-tool> edit "<prompt>" --image library/papercut/public/bg/backdrops/kitchen.jpg, so the asset prompt goes to that tool, and in edit mode a library backdrop image goes to it as a style reference; if that tool is a cloud service, that is an external send to that service, and if it is local, it stays local. Building every prop in code instead, with no image tool, is the documented fallback. (d) Voiceover: the folder says any TTS works and names only a local option, Qwen3-TTS (open weights, Qwen/Qwen3-TTS-12Hz-* on Hugging Face, running locally on Apple Silicon MPS or CUDA, usually through a small Gradio app, from my own existing install whose location the skill asks for); the narration script and, for the voice-clone variant described in style-guide.md, a reference clip and its transcript go to whatever TTS I actually choose — local with Qwen3-TTS, or to a cloud TTS if I pick one. The placeholder voice in Standard mode is local only (silent, or macOS say). (e) Fonts in a rendered project: references/remotion-recipes.md directs loading fonts through @remotion/google-fonts, so a project that follows the recipes fetches its font files from Google Fonts when it is built or rendered. The recipes also tell the agent to load other Remotion skills (remotion:remotion-best-practices and the remotion-captions skill) — separate skills, not part of this folder. - The local-only parts at use time: the read-only preflight probe in SKILL.md (sourcing $PAPERCUT_LIB/env.sh from the config file, then node -v, ffmpeg -version and Python import checks for numpy, PIL and soundfile, plus qwen_tts for Full mode) runs entirely on this machine and sends nothing — and it is a use-time probe, never part of this installation. Rebuilding the library catalogs (python3 $PAPERCUT_LIB/tools/catalog.py), the planning and writing steps, the Remotion rendering itself and the QA pass are also local, between you and me, in the series and episode files the stages read and write (research/facts.md, narration/script.json, papercut/episode.json, papercut/STORYBOARD.md, out/*.mp4). The worked example in frame-breakdown.md is invented content (a fictional product, models and numbers): reuse its event density, never its content, and never invent product behaviour — role framing is confirmed with me at intake. - Do not run ./install.sh or its --doctor, do not hand over to papercut-setup, do not run the preflight probe or any catalog rebuild, do not install Node, ffmpeg, Python packages, Remotion or Qwen3-TTS, and do not fetch my source link, generate any asset or render anything during this install: setting up the sibling skills, the library or a voice is my decision, later, when I actually ask for a reel. 3. When I say "yes, install", install it as a skill: copy the whole papercut-explainer folder into the agent's skills directory, in a folder named "papercut-explainer", keeping all 7 files (SKILL.md at its root and the six references in its references/ subfolder). A manual clone followed by copying the skills/papercut-explainer 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: papercut-explainer), all 7 files in place with matching fingerprints. Do not run any probe, script or Remotion command to verify. 5. Report what was installed, where, and what I still need to do myself (for example: if I want more than the orchestrator and its references — a full idea-to-mp4 reel — install the eight sibling skills papercut-strategy, papercut-hook, papercut-script, papercut-storyboard, papercut-build, papercut-voice, papercut-qa and papercut-setup and set up the repository's library/ myself, normally through the repository's own ./install.sh in Standard mode (Node, ffmpeg and the Python tools) or Full mode (which adds the local Qwen3-TTS voice), knowing none of that is included here; provide the source link, the one claim, the role framing, the CTA keyword and the length when I start a reel; choose my image tool and my TTS myself, knowing the use-time sends disclosed above happen only then, on the tools that run). 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.