saberongo/public-domain-books-translation/public-domain-epub-pipeline
- community
Checking that the SKILL.md file is still in the repository.
Use when creating, reviewing, or maintaining public-domain book translation projects and EPUB pipeline templates in this repository. Follow multilingual, language-pair-specific, rights-aware workflow rules.
Install
npx vskill@latest install saberongo/public-domain-books-translation/public-domain-epub-pipelineSKILL.md
Public-Domain EPUB Pipeline Skill / 公版书 EPUB 流水线 Skill
Use this skill when an AI agent is asked to create a book project, add or update a language-pair template, revise prompts, review policy documents, or build EPUB output in this repository.
当 AI agent 被要求创建书籍工程、新增或更新语言方向模板、修改 prompt、审查政策文档或构建 EPUB 时,使用本 skill。
Core Workflow / 核心流程
Canonical translation-unit gate / Canonical 翻译单元门禁
For new or migrated translation projects, treat translation_units/ as the only authoritative source-target store. Lock the full-source proper-noun discovery manifest, candidate decisions, entity_id register, occurrence ledger, terminology, and translation contract before parallel translation. When the user does not choose a proper-noun display mode, policy 3 is allowed only with selection_source=default; it never waives discovery or review. Translators write {{pn:entity_id}} templates in worker-owned chapter patches, and the single CAS merger renders names from locked CSV data. chapters/translated and chapters/final are deterministic identical-target projections and must not be edited independently.
新建或迁移翻译工程时,translation_units/ 是唯一权威 source-target store。并行翻译前必须锁定全书专名发现 manifest、候选裁决、entity_id 注册表、occurrence ledger、术语和翻译合同。用户未选择专名显示方式时可用策略 3,但必须记录 selection_source=default,且仍须完成发现与复核。译者只在自有章节 patch 中写 {{pn:entity_id}} 模板;唯一 CAS 合并器只从锁定 CSV 渲染名称。chapters/translated 和 chapters/final 是目标文本完全一致的确定性投影,不得分别手改。
For long-book parallel work, read references/adaptive_parallel_orchestration.md, run the weighted planner, and use Scheme B: disjoint translation producers, independent chapter audit consumers, and one coordinator/merger. Never infer capacity from page count alone. Unknown or unauthorized client capability means zero spawned workers. GPT-family clients may spawn at most four workers and non-GPT clients at most eight; start with no more than two and scale only after the pilot satisfies the recorded structural, proper-noun, semantic, rework, and conflict thresholds. A reviewer must not be that chapter's translation owner. Different chapters use chapter-scoped CAS and audit invalidation; same-chapter stale patches still fail.
长书并行工作必须读取 references/adaptive_parallel_orchestration.md、运行加权规划器并采用方案 B:互不重叠的 translation producer、独立 chapter audit consumer 和唯一 coordinator/merger。不得只按页数推断并发。客户端能力未知或未经用户授权时派生 worker 为 0;GPT 家族最多 4 个,非 GPT 最多 8 个;首轮最多 2 个,只有 pilot 达到结构、专名、语义、返工和冲突阈值才扩容。reviewer 不得是该章 translation owner。不同章节使用按章 CAS 和局部审计失效;同章陈旧 patch 仍必须失败。
When the active client exposes safe subagent creation, its coordinator must populate the capability file from live tool/model metadata, run npm run translation:orchestration:plan, and create no more workers than the resulting spawned_worker_count. Keep the returned task/agent IDs in run evidence and enforce the planned producer/auditor roles. Re-run the planner after the pilot, rate-limit changes, budget changes, or quality regressions. If the client cannot create workers, continue serially; do not emulate parallel writes with uncontrolled processes.
活动客户端具备安全子代理创建能力时,coordinator 必须依据实时 tool/model metadata 填写能力文件,运行 npm run translation:orchestration:plan,且实际创建数不得超过 spawned_worker_count。返回的 task/agent ID 必须进入执行证据,并严格执行规划中的 producer/auditor 角色。pilot 完成、限流变化、预算变化或质量退化后必须重跑规划器。客户端不能创建 worker 时应串行继续,不得用不受控进程模拟并行写入。
For bilingual editions, one complete source short paragraph must be followed immediately by its complete target paragraph. A unit may split one long natural paragraph only at complete-sentence boundaries; it must never cross two source natural paragraphs. Sentence-by-sentence subtitles, viewport/page chunking, or several source units followed by a target-language batch are release blockers. Structural PASS and EPUBCheck never substitute for hash-bound semantic omission/addition/neighbor-boundary audits.
双语版必须是“完整源语短段后立即跟随该段完整目标语译段”。长自然段只能在完整句群边界内部拆分,绝不跨两个源自然段。逐句字幕、按视口/页面聚块、连续多个源语单元后集中目标语都属于发布阻塞。结构 PASS 与 EPUBCheck 不能替代绑定 hash 的遗漏、增译和邻段串译语义审计。
Ordinary translation and builds use full-book static DOM/text/order/navigation checks and EPUBCheck only. A real-reader smoke test is attempted only for the final release candidate. If no supported reader is installed, record SKIPPED_UNAVAILABLE, allow release, and disclose the unverified boundary to the user; never skip the static gates.
普通翻译和构建只运行全书 DOM/文本/顺序/目录静态检查与 EPUBCheck。真实阅读器只在最终发布候选阶段尝试;未安装受支持阅读器时记录 SKIPPED_UNAVAILABLE、允许发布并向用户披露未验证边界,但不得跳过静态门禁。
- Read
AGENTS.md.
先读取 AGENTS.md。
- Before doing any repository task, read the relevant
template/files. The template is the source of truth; do not rely on memory or prior runs.
执行任何仓库任务前,必须读取相关 template/ 文件。模板是规则源,不得依赖记忆或历史执行经验。
- Read the repository README in the user's preferred language when available.
优先读取用户偏好语言对应的 README。
- Read
template/epub_pipeline/README.md.
读取 template/epub_pipeline/README.md。
- For EPUB production, cover, book-info/frontmatter, assets, quality gates, random review, or release work, read the applicable common policy files before editing or building:
template/epub_pipeline/common/README.mdtemplate/epub_pipeline/common/preproduction/stage1/_TEMPLATE.production_spec.mdtemplate/epub_pipeline/common/references/cover_design_policy.mdtemplate/epub_pipeline/common/references/book_info_frontmatter_policy.mdtemplate/epub_pipeline/common/references/epub_assets_figures_tables.mdtemplate/epub_pipeline/common/references/bilingual_parallel_edition_policy.mdtemplate/epub_pipeline/common/references/adaptive_parallel_orchestration.mdtemplate/epub_pipeline/common/references/quality_gate_framework.mdtemplate/epub_pipeline/common/references/proper_noun_display_policy.mdtemplate/epub_pipeline/common/references/note_marker_policy.mdtemplate/epub_pipeline/common/references/release_versioning.md
涉及 EPUB 制作、封面、书籍信息页/前置页、资产、质量门禁、随机评审或发布时,编辑或构建前必须读取适用的 common policy 文件:
template/epub_pipeline/common/README.mdtemplate/epub_pipeline/common/preproduction/stage1/_TEMPLATE.production_spec.mdtemplate/epub_pipeline/common/references/cover_design_policy.mdtemplate/epub_pipeline/common/references/book_info_frontmatter_policy.mdtemplate/epub_pipeline/common/references/epub_assets_figures_tables.mdtemplate/epub_pipeline/common/references/bilingual_parallel_edition_policy.mdtemplate/epub_pipeline/common/references/adaptive_parallel_orchestration.mdtemplate/epub_pipeline/common/references/quality_gate_framework.mdtemplate/epub_pipeline/common/references/proper_noun_display_policy.mdtemplate/epub_pipeline/common/references/note_marker_policy.mdtemplate/epub_pipeline/common/references/release_versioning.md
- Select the matching language-pair template, for example
French-to-English,Japanese-to-Spanish,English-to-Simplified-Chinese, orTraditional-Chinese-to-German.
选择匹配的语言方向模板,例如 French-to-English、Japanese-to-Spanish、English-to-Simplified-Chinese 或 Traditional-Chinese-to-German。
- If a matching target-language framework exists under
template/epub_pipeline/targets/{target}/, read it before translating.
如果 template/epub_pipeline/targets/{target}/ 下存在匹配的目标语言质量框架,翻译前必须读取。
- Create the book project with
books/scripts/create_book_project.py; it copiestemplate/epub_pipeline/commonintobooks/{target}/{number}_{book_id_slug}/, then overlays the matching language-pair template.
使用 books/scripts/create_book_project.py 创建书籍工程;脚本会把 template/epub_pipeline/common 复制到 books/{target}/{number}_{book_id_slug}/,再覆盖复制匹配的语言方向模板。
- Write all book-specific source text, translations, QA, EPUB output, and metadata only inside the book project.
所有具体书籍的原文、译文、QA、EPUB 输出和 metadata 只能写入书籍工程目录。
- Run the workflow through source evidence, rights checks, research, trial translation, chapter translation, review, gates, EPUB production, validation, mandatory post-EPUB stratified random spot-check, independent review, versioned release, and retrospective.
按来源证据、版权核查、研究、试译、章节翻译、审校、门禁、EPUB 制作、校验、EPUB 后强制分层随机抽检、独立评审、版本化发布和复盘流程执行。
- For translation, chapter review, random spot-check, reader feedback, and final artifact review, use
skills/expert-translation-quality/SKILL.mdto enforce expert-level prose and downstream-context polysemy back-checking without bloating every prompt.
翻译、章节审校、随机抽检、读者反馈和最终产物审阅时,使用 skills/expert-translation-quality/SKILL.md,把专家级译文与后文回看多义词复查作为执行流程,而不是把长规则塞进每个 prompt。
- Before EPUB production, run
node scripts/publication_lint.js --target={target-language} --write-reportinside the book project and fix all hard errors.
EPUB 制作前,必须在书籍工程内运行 node scripts/publication_lint.js --target={target-language} --write-report,并修复所有硬错误。 Before a translated or final chapter can pass chapter controls or template preflight, also run the structural translation coverage gate: npm run check:translation-coverage or python scripts/check_translation_coverage.py --write-report. This gate compares chapters/src/ against chapters/translated/ and chapters/final/ for severe paragraph-block shrinkage, chapter character shrinkage, lost footnote references, lost footnote definitions, and lost table/figure/formula units. A chapter control file that claims PASS cannot override this machine gate. 已译章节或终稿章节通过章节控制或模板预检前,还必须运行结构性译文覆盖门禁:npm run check:translation-coverage 或 python scripts/check_translation_coverage.py --write-report。该门禁对比 chapters/src/ 与 chapters/translated/、chapters/final/,拦截严重段落块缩水、章节字数异常缩水、脚注引用丢失、脚注定义丢失,以及表格/图片/公式单元丢失。章节 control 自称 PASS 不能覆盖这个机器门禁。
- Before final EPUB output, check long chapter titles against
references/chapter_title_policy.mdand any matching source-to-target title strategy. EPUB navigation must use concise labels, not printed-TOC title chains.
最终 EPUB 输出前,必须按 references/chapter_title_policy.md 及对应语言方向标题策略检查长章节标题。EPUB 目录应使用短题名,不应塞入纸书目录式标题链。
- After the first full-book EPUB is built, run the stratified random spot-check module from
references/stratified_random_spotcheck.mdandprompts/16a_stratified_random_spotcheck.md. Use deterministic scripts, keep all samples and evidence underbooks/{target}/{number}_{book_id_slug}/reviews/random_spotcheck/round_XXX/, close every discovered P0/P1/P2, and runnpm run review:random-validate:passbefore final output.
第一版全书 EPUB 生成后,必须执行 references/stratified_random_spotcheck.md 和 prompts/16a_stratified_random_spotcheck.md 定义的分层随机抽检模块。必须使用确定性脚本,所有样本和证据保存在 books/{target}/{number}_{book_id_slug}/reviews/random_spotcheck/round_XXX/,所有发现的 P0/P1/P2 必须关闭,最终输出前必须通过 npm run review:random-validate:pass。 The current executor must generate new random spot-check rounds. Historical PASS rounds from previous agents, previous releases, or previous private artifacts do not count. The user may specify any current-run consecutive PASS requirement of >=1; when unspecified, the default validator requires the latest consecutive 2 PASS rounds from the same review_run_id. 当前执行者必须生成新的随机抽检轮次。之前 Agent、之前 release 或之前 private artifact 已经 PASS 的历史轮次不得计入本次执行。用户可以指定任意 >=1 的当前运行连续 PASS 轮次要求;未指定时,默认强校验要求同一 review_run_id 下最新连续 2 轮 PASS。
- After random spot-check closure, run the versioned release module from
references/release_versioning.mdandprompts/18a_release_versioning.md. Createoutput/release/book_vX.X.X.epub,release_note_vX.X.X.md,release_state.json, andrelease_index.md;output/book.epubalone is not enough forDONE.
随机抽检闭环通过后,必须执行 references/release_versioning.md 和 prompts/18a_release_versioning.md 定义的版本化发布模块。必须创建 output/release/book_vX.X.X.epub、release_note_vX.X.X.md、release_state.json 和 release_index.md;只有 output/book.epub 不能标记 DONE。
- For translated titles and important proper nouns, title occurrences do not count as first body occurrences. Chapter titles, subtitles, and navigation titles must follow target-language title
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/public-domain-epub-pipeline/SKILL.md
Popularity
- Repository stars
- 262
- Repository forks
- 22
- 7-day trend score
- 28
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/saberongo/public-domain-books-translation/public-domain-epub-pipeline)