Skip to content

Short vertical cut-paper explainer reels, made by your coding agent. Nine Agent Skills plus a Remotion library. Works with Claude Code, OpenCode, Codex, Gemini CLI, Cursor, Copilot and more.

10.2 KB
Raw Blame

Prompts

Copy-paste prompts for each skill, in the order of the workflow, with what to expect. Replace anything in <angle brackets>.

Your agent picks a skill from your wording. If the wrong one loads, or none does, name it: "use papercut-hook to write 5 hooks for this fact list". Each skill reads the files the stage before it wrote, so you can stop after any stage and pick up in a new session. examples/paper-trail/ shows what every stage writes.

End to end

Make a papercut explainer about <link>, 25 seconds, CTA keyword PAPER
  • Loads papercut-explainer, the router. It checks your install mode, finds $PAPERCUT_LIB and your series folder, then asks which run mode you want (full reel, script only, storyboard from a script, style prompt, fix one scene).
  • Asks only what is missing: the claim the video proves, who does the work in the product (role framing), what the viewer gets for the keyword, and the audience level if there is no SERIES.md.
  • Writes the BRIEF at the top of episodes/NN-slug/research/facts.md, then hands to papercut-strategy and walks the stages. It renders only when you ask.

Other ways in:

Write only the script, banners and staging table for <link>. No video yet.
Storyboard this script: <path to narration/script.json>
Give me the papercut style prompt for <image or video tool>

These pick the "script only", "storyboard from an existing script" and "style prompt" run modes. The last one hands you a block from references/style-prompt.md and stops.

1. Setup

Set up papercut-explainer, partial install
  • Loads papercut-setup. It runs ./install.sh --doctor from the repo root and tells you in plain words what is installed and which mode that adds up to.
  • Asks, in one question: how much (Minimal, Standard, Full or Custom), skills for every project or one, symlink or copy. "Partial" usually means Minimal, or Custom with the components you name (skills,deps,pytools,verify). It asks again before any multi-GB download.
  • Runs install.sh --yes with the matching flags and writes ~/.config/papercut-explainer/config.env, which every later skill reads for the library path. A new Claude Code, Codex or Gemini CLI session picks up the linked skills (restart it); opencode reloads them automatically.

2. Strategy

Pick the topic, format and funnel for a reel about <link>. Audience: <who>.
  • Loads papercut-strategy. It reads SERIES.md first, and if there is none it asks for the audience (who, level, regions), voice, palette, mascot and keyword log, then writes it.
  • Picks the one familiar format for the whole reel ("what show is this secretly about?"), the mode, the keyword and the freebie, and plans the caption.
  • Writes SERIES.md and the BRIEF block at the top of research/facts.md. Next stage: the hook.

The same skill also runs competitor studies and the post-publish autopsy:

Study these 4 reels and tell me what to borrow: <url> <url> <url> <url>
Here are the numbers for the reel I posted: <views, skip rate, watch time, comments>. Do the autopsy.

The first spawns one subagent per reel and saves research/analytics/YYYY-MM-DD-<creator>-reels.md. The second reads the numbers hook first, topic second, and writes research/analytics/YYYY-MM-DD-autopsy.md plus an update to SERIES.md.

3. Hook

Write 5 hooks for this fact list: <paste the facts, or point at research/facts.md>
  • Loads papercut-hook. It needs the BRIEF. If there is none it sends you to papercut-strategy.
  • Writes five variants from at least three hook patterns, scores each on the six-question hook check, drops any that fail the facts, and picks the best.
  • Writes the HOOK block (SAY, WRITE, SHOW, ALIGN) at the top of papercut/STORYBOARD.md. The WRITE line becomes the hook title in episode.json when papercut-build fills it in. It also sets the rule for the motion out of the hook, which the storyboard applies.

4. Script

Write the script, banner track and staging table for <link>
  • Loads papercut-script. It needs the BRIEF and the HOOK block. It researches the source (web fetch or search, a browser for screenshots) and asks you for screenshots and facts if it has none of those.
  • Writes the sourced facts list under the BRIEF in research/facts.md, the 75-95 word script on the 10-beat spine in narration/script.json (line l1_hook is the SAY line), 10-18 banners, and a staging table with one set per banner and a GAG column from the humour pass.
  • The staging table goes into papercut/STORYBOARD.md under the HOOK block. Stressed words per line are marked there for the voice stage.

5. Storyboard

Storyboard this script
  • Loads papercut-storyboard. It needs narration/script.json and the HOOK block and staging table. It reads the two library catalogs to check every camera move, cut kind, backdrop and SFX name.
  • Writes the global rules, the Camera / Cut table (one row per scene, no plain CameraPush on more than a third), one block per scene with an event on each stressed word, and the SFX plan into papercut/STORYBOARD.md.
  • The cut column becomes cuts in episode.json at the next stage. The Timing section is left for the voice stage.

6. Voice

Three paths. Say which one, or the skill asks.

A. Local Qwen3-TTS (Full install)

Voice the episode with local Qwen3-TTS, speaker Ryan
  • Loads papercut-voice. It checks that qwen_tts imports and hands to papercut-setup if not. It copies voice.example.json to voice/voice.json and sets the speaker and the style line (instruct).
  • Plays the first three lines before the full run, then generates takes, picks one per line and writes voice/final_narrator/<id>.wav and timing_narrator.js.
  • Finishes with build_vo.py and timing_md.py: public/vo/captions.json, src/plan.json and the Timing section of STORYBOARD.md.

B. Your own recording or another TTS

I recorded the narration myself, one file per line, in <folder>. Time the script to it.
  • Loads papercut-voice, path B. It needs an ASR backend (mlx-whisper or faster-whisper) for word times, and asks for one WAV per line or raw takes.
  • Writes voice/final_narrator/, the ASR file and timing_narrator.js, then the same captions, plan and Timing section.
  • Without ASR it can only estimate 2.75 words a second, and says that captions will drift.

C. Placeholder voice

Use a placeholder voice so I can see the timing
  • Loads papercut-voice, path C (Standard install). It runs placeholder_vo.py from narration/ with a pace of 3.5 to 4 words a second, silent or the macOS say voice.
  • Writes silent WAVs and timing_narrator.js, then runs build_vo.py, which flags any scene under 1.2 s. You can build, preview and QA the whole reel, and re-voice later without touching a scene.

7. Build

Build scenes S03-S06 with parallel scene agents
  • Loads papercut-build (and the Remotion best-practices skill). It needs papercut/STORYBOARD.md and src/plan.json. It builds S01Hook itself first, as the reference scene.
  • Splits S03 to S06 across scene agents, two or three scenes each, and gives each the project path, its frame ranges, a QA folder and library/papercut/AGENT_BRIEF.md. Each agent writes only src/scenes/<Scene>.tsx and tools/cue_sheet.d/<Scene>.json.
  • Agents QA their own stills. You get a short report per agent. Nothing renders unless you ask.

Other build requests:

Scaffold the episode as 01-<slug> in <series folder>
Mix the sound for the episode
Render it

They run new_episode.sh (and merge your earlier files into the new folder), then cues.py with audio/render.mjs, then render.sh for out/<name>.mp4.

8. QA

QA the reel
  • Loads papercut-qa. It renders contact sheets with qa.sh, four to six frames per scene, one render at a time.
  • Checks them against the style checklist and the storyboard: paper look, banner rules, the HOOK block on screen, a camera move in every scene, humour pass, bottom 20% clear.
  • Fixes go back to papercut-build, at most two rounds per scene. The last step is a delivery summary that names what is still placeholder (voice, music, SFX, assets).

Maintenance

Re-voice with a different speaker

Re-voice the episode with a different speaker

Loads papercut-voice. It asks which speaker (Aiden is the other recommended one), changes speaker in voice/voice.json, regenerates the takes, finalizes them, then reruns build_vo.py and timing_md.py. Scene starts move with the new word times, and the scene cache re-renders everything.

Swap the mascot's colour

Swap the mascot's colour to <colour>

Loads papercut-build. Say whether the colour is for this episode or the whole series. For one episode it changes the color and shade props the scenes pass to the Critter, and only those scenes re-render. For the series it changes the hero colour in the library's kit/tokens.ts, re-syncs with sync_kit.sh (a kit change re-renders everything) and notes the colour in SERIES.md. It never edits src/lib/ inside a project, because that folder is a synced copy.

Add a backdrop

Add a backdrop for <place>

Loads papercut-build. It checks the 24 existing sets first, and generates or draws a new one only if none fits (an image tool, or code like backdrops-src/). The JPG goes in library/papercut/public/bg/backdrops/, with an entry in backdrops.json and the BACKDROPS table in kit/paper/backdrops.tsx. Then it runs tools/catalog.py and sync_kit.sh.

Add a sound effect

Add a sound effect: <describe it>

Loads papercut-build. It writes library/audio/sfx/<name>.js with the // @sfx name | tags | seconds | description | origin header line, runs tools/catalog.py so it shows in audio/CATALOG.md, and adds a row to the scene's tools/cue_sheet.d/<Scene>.json. Then cues.py and render.mjs rebuild the audio.

Fix scene S04 so the camera moves

Fix scene S04 so the camera moves

Loads papercut-explainer (run mode "fix one scene"), which hands to papercut-build and then papercut-qa. It reads S04's row in the Camera / Cut table. If the table names no camera move for S04, run papercut-storyboard first to choose one. Then it edits src/scenes/S04*.tsx to use that move, checks the scene with qa.sh, and re-renders only that scene.