| name | translate-skill |
| description | 将 mattpocock/skills 的内容翻译、刷新或复核到简体中文本地化仓库 vinvcn/mattpocock-skills-zh-CN 时使用这个项目级 skill。适用于 skill files、README content、CLAUDE.md、CONTEXT.md、docs，以及其他需要保留行为关键 identifiers 的上游用户可见内容。 |
Translate Skill
Use this skill to translate upstream mattpocock/skills content into Simplified Chinese for vinvcn/mattpocock-skills-zh-CN.
This skill is for content localization, not Git synchronization.
Operating principle
Translate user-facing English prose into natural Simplified Chinese while preserving all behavior-critical content exactly.
The target repo is an independent Simplified Chinese localized edition. It should receive translated content, not upstream repository metadata.
What to translate
Translate natural-language prose, including:
README explanations
skill instructions
skill descriptions
user-facing frontmatter prompts
agent-facing guidance
maintainer-facing guidance
docs prose
examples that are written as prose
What to preserve exactly
Do not translate or rewrite:
directory names
skill names
slash commands
CLI commands
code blocks
inline code
file paths
package names
tool identifiers
API identifiers
environment variable names
frontmatter keys
JSON/YAML/TOML keys
Markdown link URLs
behavior-critical labels
Preserve Markdown structure, heading levels, list nesting, tables, link targets, relative paths, and code fences.
Repository-path localization
When translating user-facing installation examples, replace the upstream repo path:
mattpocock/skills
with the localized repo path:
vinvcn/mattpocock-skills-zh-CN
Only make this replacement where the command or prose is telling users how to install or use the localized repo.
Do not remove attribution to the upstream project.
Frontmatter rules
Preserve frontmatter keys exactly.
Classify each frontmatter value by meaning, not just by field name:
- Keep
name values unchanged.
- Translate user-facing or agent-facing natural-language prose into Simplified Chinese. This
includes
description and argument-hint values.
- Keep identifiers, commands, paths, package names, URLs, tool names, booleans, numbers, and
other typed configuration values unchanged.
- Within a translatable value, preserve embedded slash commands, inline code, placeholders,
paths, URLs, package names, tool names, and other behavior-critical spans exactly.
- Flag ambiguous values rather than guessing.
Example:
---
name: teach
description: 在这个工作区中教用户一个新技能或概念。
disable-model-invocation: true
argument-hint: "你想学习什么？"
---
Translation style
Use Simplified Chinese that is:
natural
developer-friendly
concise
accurate
consistent with existing repo tone
Keep common engineering terms in English when the English term is standard among developers or when translating it would reduce clarity.
Workflow for a single file
When translating one file:
- Identify the file path and file type.
- Classify the file as translatable prose, mixed content, config/metadata, or non-translatable.
- Protect behavior-critical spans before translation.
- Translate only natural-language prose.
- Restore protected spans exactly.
- Check that commands, code blocks, paths, URLs, identifiers, and frontmatter keys are unchanged.
- Check that localized install commands use
vinvcn/mattpocock-skills-zh-CN.
- Return the translated file content or a patch, plus any review flags.
Workflow for a repo refresh
When refreshing from upstream:
- Treat upstream as a content source, not as Git history.
- Identify new, changed, and removed content files.
- Translate new and changed prose-bearing files.
- Copy or preserve non-translatable support files only when they are in scope.
- Preserve the localized repo’s README positioning and install path.
- 按“验证步骤”完成同步后的结构、完整性、索引和行为关键内容检查。
- Add a concise entry to the README sync log.
- 在 README 中记录本次验证结果、翻译执行者和翻译策略摘要。
- Flag ambiguous files, removed files, or risky transformations for maintainer review.
- 总结已翻译文件、复制或保留文件、移除文件、跳过文件、验证结果和复核标记。
验证步骤
每次上游内容刷新后，必须完成并记录以下检查：
- 运行
node scripts/check-translation.mjs，确认 Markdown 结构、frontmatter、README install path 和 license invariant 没被破坏。
- 检查公开 skill 索引一致性：
engineering/、productivity/、misc/ 下的 skills 必须同时出现在顶层 README.md 和 .claude-plugin/plugin.json；personal/、in-progress/、deprecated/ 不应出现在 plugin 或顶层公开索引中。
- 对比
upstream/main 的 in-scope 文件清单，确认没有缺失上游文件，也没有保留已经从上游移除且不属于本地策略的 stale files。
- 检查共同 Markdown 文件的行为关键内容：frontmatter keys 和
name 值不变，fenced code blocks 平衡，路径、命令、URL、identifier 不被误改。
- 运行
git diff --check 和 git diff --cached --check，确认没有 whitespace 或 patch hygiene 问题。
- 检查 README 同步记录指向最新 upstream short SHA，并包含本地同步 commit；不能留下“待定”占位。
- 扫描 stale install path 和旧路径，例如仍指向上游 repo 的安装命令、旧的中文仓库短路径、已移除的 triage skill 名、已移除的 domain-model 相对路径等。
- 运行
node scripts/audit-english.mjs 作为人工复核队列。该脚本在本仓库会包含大量合理的英文术语、命令、示例和 identifiers，因此只作为 review flag，不作为硬性失败门槛。
验证结果应同步写入顶层 README.md，使用简短 checklist，不要把完整命令输出粘进去。
README sync log
Each upstream refresh should update the top-level README.md sync log with one short entry.
The entry should include:
- the refresh date in
YYYY-MM-DD format
- the upstream source revision, usually
mattpocock/skills@<short-sha>
- the local sync commit, if it already exists; otherwise use a short pending note and replace it after commit
- a one-sentence description of the visible content change
Keep this record concise. Do not move the detailed translation workflow into the README; link to this skill instead.
Example:
- 2026-05-09: Synced upstream `mattpocock/skills@733d312`, local commit `c9fe120`. Added Chinese translations for `prototype` and `in-progress` content, and refreshed public skill indexes.
Required output format for review
When reviewing a translation refresh, provide:
Changed files:
- ...
Translated files:
- ...
Copied or preserved files:
- ...
Removed or stale files:
- ...
Review flags:
- ...
README sync log:
- ...
验证结果:
- ...
Invariant checks:
- install commands point to vinvcn/mattpocock-skills-zh-CN
- code blocks preserved
- frontmatter keys preserved
- paths and identifiers preserved
- Markdown structure preserved
Fail-closed rule
When unsure whether text is behavior-critical, preserve it and flag it.
Do not silently rewrite anything that could affect installation, skill discovery, command execution, file references, API calls, tool use, or agent behavior.
