shuohao-skills — Adelie plugin

An Adelie plugin packaging shuohao-skills — agent-agnostic skills for AI short-drama production — so a Project can install them from the plugin library.

The pipeline turns a novel into shoot-ready material. The outline converges the structure; script, scenes and characters iterate together; the storyboard only outputs, it makes no new decisions.

| Skill | What it does | | --- | --- | | novel-outline | Adapts a novel into a five-piece short-drama outline: adaptation notes, cast, beats, per-episode synopses, asset list. All 14 quality gates are script-checked; a checkup mode diagnoses an existing outline | | novel-characters | Turns the cast into a character bible: profiles, design prompts, voice prompts, model sheets. Report language is configurable (--lang) | | character-refs | Actually generates reference images for characters from any story: a front full-body anchor first, every other view referencing only that anchor, generated on demand in tiers. Backends: Qwen Image (ComfyUI), codex built-in image generation, the OpenAI Images API (GPT Image 2) or a custom command | | novel-art | Art bibles for AI production (scenes + narrative props): consistency anchors, lighting and state variants, scale references, no-people/no-hands white plates. All 10 quality gates script-checked | | novel-script | Screenwriting for AI short drama: scenes + beat flow, per-episode duration estimated from reading speed, a gated cold-open hook, a per-character line book with voice prompts. All 10 quality gates script-checked | | novel-storyboard | Storyboarding: segments (≤15s) → cuts (2–5s hard gate) → keyframes, with MiniMax H3 prompt alignment and cut times audited verbatim. All 18 quality gates script-checked |

The five pipeline skills render their reports in English too — pass --lang en to render (report language defaults to Chinese, and data content stays as authored). character-refs ships Chinese, English and Japanese UIs. The five stage reports can be merged into one page with a left-hand nav:

node scripts/report.mjs --from <demo-dir> --out report.html
node scripts/report-selftest.mjs   # 92 assertions, no browser needed

Install

From the plugin library. In Adelie, open the plugin library and upload shuohao-skills-<plugin version>.zip — the archive's top-level directory names the plugin. It lands in the data root's plugins/ directory and is then enabled per agent.

From npm. The package holds the same tree, but install from the zip rather than from the npm tarball: npm wraps the bundled files whose names are Chinese (渡口-outline.json, 阿禾-intake.json) in PAX tar headers, and Adelie's tarball reader does not decode those, so a tarball install fails on those names. The zip route keeps them intact.

npm view @lmliheng/shuohao-skills
npm pack @lmliheng/shuohao-skills     # the same tree, to inspect or repackage as a zip

Requirements

| | Required? | Notes | | --- | --- | --- | | Node | Yes | ≥ 18. The scripts use only the standard library — no npm dependencies | | Model quota | Yes | Uses the current session's quota. No API key needed | | Image model | Only for character-refs | One of: your own ComfyUI (Qwen Image), codex built-in image generation, an OpenAI API key (GPT Image 2), or a custom command |

Output language: the skills are Chinese-first, as upstream — novel-characters writes Chinese character profiles even for an English source novel. See skills/novel-characters/README.en.md.

Licence and credits

The skills are the work of 烁皓 (eternityspring) — upstream: https://github.com/eternityspring/shuohao-skills, licensed under Apache-2.0 (LICENSE, NOTICE).

This plugin only repackages them for Adelie: the skill directories and the repo-level report assembler are copied verbatim, and each SKILL.md frontmatter is rewritten into the Penguin shape (single-line keys, short_description / short_description_zh, a dated version) because the Penguin frontmatter parser does not read YAML block scalars or nested maps. The upstream scripts/install.sh (a symlink installer for Claude Code and Codex) is not part of the plugin.