Design & Motion · JimLiu/baoyu-skills

baoyu-comic

Name: baoyu-comic
Author: JimLiu/baoyu-skills

Knowledge comic creator supporting multiple art styles and tones. Creates original educational comics with detailed panel layouts and batch-capable image generation. Use when user asks to create "知识漫画", "教育漫画", "biography comic", "tutorial comic", or "Logicomix-style comic".

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandWrite FileDual-pass Audit

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Perform "deep content analysis" before drawing: Produce analysis.md following the analytical framework (audience, value proposition, core themes, characters & story arcs, content signals), and recommend art/tone/layout and page count based on these signals.

skills/baoyu-comic/references/analysis-framework.md

Deep analysis framework for transforming source content into effective visual storytelling.

Page counts are tiered according to content complexity (short 5-8 pages, medium 9-15 pages, full biography 16-25 pages), and recommended values are recorded in the front matter of analysis.md for subsequent steps to reference.

Style operates in three dimensions—art(6) × tone(7) × layout(7)—plus a preset layer: Presets are not just aliases for art+tone, but also enforce additional rules (e.g., ohmsha bans "talking heads" and mandates visual metaphors).

skills/baoyu-comic/references/workflow.md

**Preset vs Art+Tone**: Presets include special rules beyond art+tone.

The four directories art-styles/, tones/, layouts/, and presets/ load definition files on demand; the ohmsha preset also pre-configures default characters like Nobita/Doraemon, forbidding ad-hoc character creation if unspecified by the user.

Character consistency is engineered via "character sheet + reference image": First generate characters/characters.png to serve as a cross-page --ref; if the backend does not support --ref, it falls back to embedding character descriptions into each page's prompt.

skills/baoyu-comic/SKILL.md

| Exists | Supported | Pass sheet as `--ref` on every page |

Includes a failure recovery chain: compress reference images with sips/pngquant to avoid payload issues → retry → if it still fails, drop --ref and fall back to textual embedding.

Page-by-page generation follows a strict dependency order and batching strategy: Produce the character sheet first, then dispatch pages in batches; each page's prompt must be persisted to disk before invoking the backend.

skills/baoyu-comic/SKILL.md

Honor workflow dependencies first: generate `characters/characters.png` before pages that use it as a reference.

The section explicitly mandates: Never start the first page batch until all selected page prompt files exist on disk; batch failures only retry the failed items.

Includes an executable script to merge pages into a PDF: Run merge-to-pdf.ts via bun, build the document with pdf-lib according to actual image dimensions, and save to disk via Bun.write.

skills/baoyu-comic/scripts/merge-to-pdf.ts

const pdfDoc = await PDFDocument.create();

The script sorts by the NN filename prefix (00-cover-*.png, 01-page-*.png …) and exits with an error if pages are missing; default output is <dir>/<dirName>.pdf.

Supports partial workflows to avoid full re-runs: Generate only storyboards, only prompts, only images, or regenerate specified pages.

skills/baoyu-comic/SKILL.md

| `--regenerate N` | Regenerate specific page(s) only (e.g., `3` or `2,5,8`) |

Used in combination with --storyboard-only/--prompts-only/--images-only; details are documented in references/partial-workflows.md.

Mandatory backup before regeneration: Existing prompts and images are renamed with a -backup-YYYYMMDD-HHMMSS suffix to preserve comparable historical versions.

skills/baoyu-comic/references/workflow.md

**Backup rule**: If `characters/characters.png` exists, rename to `characters/characters-backup-YYYYMMDD-HHMMSS.png`

The same rule applies to pages and prompts; generation failures or retries do not overwrite original files.

Interaction language is maintained throughout the pipeline: Storyboards, prompts, options, and progress updates all use the user's input language or saved preferred language.

skills/baoyu-comic/SKILL.md

**Rule**: Use user's input language or saved language preference for ALL interactions:

Technical terms remain in English; language priority order is --lang > EXTEND.md > conversation language > source material language.

2Core Capabilities

01Deep content analysis → analysis.md (audience/value/theme/character arc/content signals)

02Storyboard + character design (storyboard.md, characters/characters.md), allowing review before image generation

03art × tone × layout × preset selection and automated recommendation (including ohmsha/wuxia/shoujo/concept-story/four-panel)

04Cross-page character consistency (character sheet as --ref / textual description fallback)

05Batch page generation + single-item retries + dependency ordering (character sheet precedes pages)

06Local PDF merging (bun + pdf-lib, sorted by NN prefix)

07Partial re-runs (--images-only / --prompts-only / --storyboard-only / --regenerate N) with backup rules

08Artifact compression to fit --ref payloads (sips / pngquant)

3External Dependencies

Kind	Dependency / Endpoint
cli	bun / npx（${BUN_X} 运行时，跑 scripts/merge-to-pdf.ts）
package	pdf-lib（脚本内 import，用于建 PDF）
cli	sips / pngquant（压缩角色表，避免 --ref payload 失败）
cli	codex CLI / baoyu-image-gen（图像生成后端，codex-imagegen 分支）
api	Codex 原生 imagegen / Cursor GenerateImage / Hermes image_generate（宿主原生图像工具）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Graded/per-use authorization, sandbox recommended

For Administrators

Approval workflow, rollback support, full audit

Evidence & Triggers (File Paths & Excerpts)

scripts/: Executes local scripts or commands: scripts/merge-to-pdf.ts (run with bun; reads local page images only, writes PDF)
本地工作目录: File I/O and artifact output: Under comic/{topic-slug}/: source-{slug}, analysis.md, storyboard.md, characters/ch

Audit Notes: (a) SKILL.md specifies external image generation backends: '**Codex (`imagegen`)** — first, inspect your available-skills / tool inventory.' and 'route through `baoyu-image-gen --provider codex-cli` (preferred)'; following SKILL.md transmits user source text (biographies/tutorials) and entire storyboard prompts outbound to third-party image models, and multi-page batching substantially consumes user account quotas. (b) Heavy local disk writes: materials, analysis, storyboard, characters, prompts, page images, PDFs under comic/, as well as EXTEND.md. (c) The bundled script scripts/merge-to-pdf.ts is purely local (builds PDF via pdf-lib + Bun.write) with no network or credential access, fitting blue criteria; red tier criteria (reverse-engineered APIs/client impersonation/bypassing protections/arbitrary code execution) are not met. Following precedents like ai-image-generation / seedance, 'sending source materials to generation backends' is consistently categorized as orange.

5Second-Pass Verification

[ok] 自带脚本只有 merge-to-pdf.ts，且仅本地行为 — 逐行核对：import fs/path/pdf-lib，createPdf() 里 readFileSync 页面图 → embedPng/embedJpg → Bun.write(outputPath)；无网络调用、无 env 读取、无子进程。
[ok] pdf-lib 依赖来源（无自带 package.json） — skill 目录确无 package.json；依赖靠仓库根 workspace 的 dependencies（pdf-lib ^1.17.1）或运行环境解析。这一点写进 risks，不影响档位。
[ok] SKILL.md 点名单一生成后端（orange 依据） — Codex imagegen / codex CLI (baoyu-image-gen --provider codex-cli) / Cursor GenerateImage / Hermes image_generate 均在 SKILL.md 正文，且 7.2 要求“生成任何页面前先读后端 SKILL.md 判断 --ref 能力”。
[ok] 角色一致性三策略与失败恢复 — SKILL.md 7.2 表（Exists/Supported、Exists/Not supported、Skipped）与 workflow.md 的 Strategy A/B/C 一一对应，另有 --ref 失败时的压缩→重试→降级链。
[ok] 预设与默认角色 — workflow.md 列出 ohmsha 默认四角色（大雄/哆啦A梦/胖虎/静香）并写明不得为 ohmsha 自造角色；presets/ 5 个文件均在盘。
[ok] 备份与局部重跑规则 — workflow.md 7.1 备份规则、SKILL.md Partial Workflow Options 表与 partial-workflows.md 均存在。
[ok] meta 元数据 — GitHub API：MIT、25926 stars、pushed_at 2026-09-10T15:13:43Z；本地 HEAD 与任务 pin 1567581c26ec… 一致。

6Verdict & Scenarios

Broad yet controllable style space: 6 art × 7 tone × 7 layout × 5 presets, with presets enforcing extra narrative rules (e.g., ohmsha's metaphor requirement, four-panel's Kishōtenketsu structure).

Long-form project friendly: Multiple review gates + partial re-runs + backup rules allow long comics to be delivered in stages and rolled back.

Engineered character consistency: Character sheets act as reusable reference assets across pages, with built-in fallbacks for backends lacking --ref support.

Complete deliverables: Instead of scattered images, an included script merges them into a PDF in NN order, ready for distribution.

Front-loaded content analysis framework avoids loose narrative caused by rushing straight into drawing.

Best for: Individuals or teams adapting long-form articles, biographies, or tutorials into educational comics, especially those needing cross-page character consistency and deliverable PDFs; also serves as a reference for skill authors studying storyboard → prompt → image pipeline design.

Not for: Not suitable for single standalone illustrations (use baoyu-article-illustrator instead); not suitable for strictly offline environments or where data egress is forbidden; not suitable for rapid single-pass drafts that bypass multi-stage approvals.

Install Agent-Ready

① Verified Mirror 2026-09-15

Option A · Direct DownloadDownload baoyu-comic.tar.gz

sha256: 0b7b74a684ebd49d…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 1567581c26.