Workflow & Automation
explainer-video-from-coursework
akshayDev17/ai-explainer-video · ★ 0
Turn a directory of coursework notes into a narrated, animated explainer-video series — an agent skill for Claude Code, Codex, DeepSeek Harness and Antigravity.
Manifest details
安装
dsh plugin --profile web add 'explainer-video-from-coursework@0.4.0'从 npm 安装预构建包
For install and environment setup, plus plugin development tutorials, see DSH community docs ↗
README
Table of contents
Turn a directory of structured coursework notes into a narrated, animated explainer-video series.
A skill for AI coding agents — Claude Code, Codex, DeepSeek Harness and Antigravity. Point it at a folder of course notes: it plans the series, writes the narration, synthesizes the audio, authors the visuals, then renders and assembles finished MP4s with YouTube-ready chapters and captions.
Its planner does not merely invent titles — it proves completeness: every
content-bearing section is assigned to a video or classified as navigation
(0 orphaned) before any expensive stage runs.
The pipeline
Stage 0 (/inputs) collects the two things only you can decide — the source
notes directory, and where the finished videos land.
| # | stage | what it does | cost |
|---|---|---|---|
| 1 | /plan |
decompose the notes, prove coverage | free, local |
| 2 | /enrich |
retitle and curate (skip / merge) | free, agent |
| 3 | /script |
write the narration segments | free, agent |
| 4 | /audio |
TTS → a measured timeline | network — spends the Gemini key |
| 5 | /scene |
author the visuals (diagram / code / svg) |
free, agent |
| 6 | /publish |
YouTube chapters + captions | free — no TTS, no Chromium |
| 7 | /render |
frames → H.264 MP4 | local CPU |
| 8 | /assemble |
per-segment parts → the finished cut | local CPU |
Stages 1–2 run once for the whole directory; stages 3–8 run once per video. One video end to end is a valid stopping point — a pilot, or a single-target run.
Install
Claude Code
claude plugin marketplace add akshayDev17/ai-explainer-video
claude plugin install explainer-video-from-coursework@akshaydev17
Codex
codex plugin marketplace add akshayDev17/ai-explainer-video
codex plugin add explainer-video-from-coursework@akshaydev17
DeepSeek Harness
dsh plugin --profile web add explainer-video-from-coursework
Antigravity
agy plugin install https://github.com/akshayDev17/ai-explainer-video
All four commands install the same npm package — one package, three manifest surfaces,
one copy of all sixteen skills: the ten video skills, coursework-author, the companion
skill for writing the coursework these videos come from, and the five code-flow skills:
explainer-video-from-codebase, which makes explainer videos of a codebase's major code flows
(Python, TypeScript / JavaScript, Go, C, C++, or a mix), and its four stage skills. It mines the
repo, ranks candidate flows, optionally traces the test suite in a sandbox, has you approve the
flow list, plans the videos, then runs the same per-video stages with verified code panels and
generated flow diagrams. How each platform reads it, and how
to publish, is in
DISTRIBUTION.md.
Prerequisites
- Node ≥ 22 —
render.mjsrelies on the built-in globalWebSocket. ffmpeg+ffprobe— H.264 encoding, muxing, and duration measurement.- Headless Chromium —
npx playwright install --with-deps --only-shell. - A Gemini API key — for TTS. It lives in the OS secret store (macOS
Keychain · Linux
secret-tool· Windows Credential Manager) under the service namegemini-tts— never in env, never in a plaintext file. A free key: https://aistudio.google.com/apikey.
The tools fail loud on a missing prerequisite rather than quietly producing a
broken video. Per-OS commands and citations:
SETUP.md.
Use it
Invoke the skill with the notes directory:
/explainer-video-from-coursework path/to/your/notes/
You get one .mp4 per video, plus description.md (blurb and a paste-ready
chapter block) and captions.srt, copied to the destination you chose — by
default videos/ beside the source. Scratch stays in a gitignored
out/<name>/<NN>/.
For a codebase, invoke the code-flow skill with the repo root instead:
/explainer-video-from-codebase path/to/your/repo/
It shows you the ranked candidate flows and its proposed series, and plans no video until you approve the flow list. Its scratch root sits outside the repo.
Run every command from the project (where out/ lives), not from the skill's
install folder. The tools self-locate their own tools/, src/ and scenes/,
so only the work and scratch paths depend on the current directory.
Repository layout
plugin.json Agent Plugins manifest (Codex, Antigravity)
.claude-plugin/ plugin.json + marketplace.json (Claude Code)
.agents/plugins/ marketplace.json (Codex)
cordis.patch.yml + lib/index.js DSH bundle plugin (serves this package's skills/)
skills/explainer-video-from-coursework/ the main skill: SKILL.md, tools/, src/, scenes/, SETUP.md
skills/craft-video-<stage>/ the 9 stage skills, separately invocable
skills/coursework-author/ the companion skill: authoring the coursework (vendored copy)
skills/explainer-video-from-codebase/ the code-flow skill: SKILL.md, FLOWS.md, tools/, vendored tree-sitter
skills/craft-codeflow-<stage>/ its 4 stage skills (mine, flows, plan, scene), separately invocable
DISTRIBUTION.md publish + install commands
Docs
| file | what it covers |
|---|---|
SKILL.md |
the contract of record — the agent's entry point |
SKILL.md (codebase) |
the code-flow pipeline's contract of record |
FLOWS.md |
every code-flow artifact's fields, from the code map to the code panels |
SETUP.md |
cited, per-OS machine prerequisites |
ARCHITECTURE.md |
how the pieces fit together |
PIPELINE.md |
the stage-by-stage pipeline |
ROADMAP.md |
design record and open work |
DISTRIBUTION.md |
publishing, and installing on each platform |
RELEASING.md |
how a release works — a dispatch, not a merge |
License
MIT — see LICENSE.