| name | project-onboarding |
| description | Read-only orientation for a new agent session in a learning project. Use when the host prompts "请使用 project-onboarding skill 了解这个项目". Read project state, summarize in markdown, do NOT modify any files. Run once per session (typically at init). |
Project Onboarding
Orient yourself in a learning project workspace. Read-only. Do not modify files.
When invoked
Host prompt will include the project path. If $ARGUMENTS.project_path is given, use it; otherwise the agent SDK's cwd is the project root.
Steps
Use Read and Glob tools only. No Write, no Edit, no Bash.
- Read
{project_path}/.learning/project.json. Expect fields: name, created, chapters[], total_duration. If missing, the project hasn't been initialized — tell the user and stop.
- Glob
{project_path}/*.md to find chapter files (zero-padded, e.g. 00-*.md, 01-*.md). Sort by name to get generation order.
- Read the most recent chapter file (last in sorted order). This is the current progress anchor and shows the writing style to maintain.
- Read the most recent
*.concepts.json to extract top concepts. If none, the project is fresh.
Output
Return a single markdown text response (the host displays it as the agent's first message):
## 项目摘要
**主题**: {project.name}
**进度**: 已完成 {N}/{M} 章 ({percent}%)
**当前焦点**: {latest chapter title or "尚未开始"}
**已学概念**: {top 3-5 from concepts.json, comma-separated; "无" if fresh}
**接下来**: {one-line suggestion: 完成第 N+1 章 / 复习 / 等等}
Keep it under 200 words. No code blocks, no Mermaid.
Failure modes
project.json missing → respond 项目尚未初始化 and exit- No
*.md files yet → say 已规划 {M} 章但尚未生成任何内容
- Permission errors reading files → list which files failed; don't silently skip
