name	msd-course-maintenance
description	Maintain the Modern Software Development course lab manual repository. Use when revising MSD chapters, README indexes, instruction.md manuals, grading documents, comprehensive AI Agent labs, cross-platform run-through notes, or repository-facing teaching guidance for this course.

MSD Course Maintenance

Purpose

Use this skill when maintaining the MSD repository for the Modern Software Development lab manual. The repository is a student-facing course manual, not a product site or internal engineering wiki. Changes should help students complete experiments independently on their own laptops.

Repository Shape

README.md is the course entry and topic index.
Chapter folders use sortable prefixes: 00_0, 00_1, 01 through 10, and 99.
Most chapters contain instruction.md, 实验报告提交及评分标准.md, and 附件/.
instruction.md is the experiment manual. 实验报告提交及评分标准.md is the submission and grading contract.
Keep chapter titles, README links, folder names, and grading document titles synchronized.

Teaching Assumptions

Students usually have one personal laptop.
The course has no teaching assistants and no required classroom demo session.
Documents must be self-contained enough for students to work during lab time or finish later.
Students may use AI tools for troubleshooting; unresolved problems should be discussed with the 授课老师.
Avoid references to asking assistants, receiving demo-only materials, or relying on resources outside the manual unless the user explicitly adds that teaching process.

Editing Rules

Read README.md and the target chapter before editing.
Keep edits scoped to the requested chapter, index entry, or grading document.
Keep student-facing prose in a teacher-authored Chinese style: use 同学们, 本实验, 本阶段, 授课老师.
Do not write real server addresses, accounts, tokens, API keys, internal endpoints, or private connection details into the repository.
When a manual changes required artifacts, logs, tests, model usage, environment paths, or safety requirements, update the matching grading document.
Do not put detailed submission rubrics in instruction.md when a separate grading document already carries them.
Keep chapter boundaries clear. Do not re-teach earlier chapters except as prerequisites or short reminders.

Course-Specific Priorities

Prefer Windows-laptop-friendly instructions when the audience is students; add macOS/Linux differences when they matter.
Prefer conda as the required local environment path. Docker may be recommended or used for reproducibility when feasible.
For LLM and Agent chapters, require real model evidence when the experiment says it is mandatory. Fake clients are for deterministic tests and offline troubleshooting only.
For Agent topics, make tool contracts, MCP-style interfaces, logs, failure paths, safety boundaries, and verification evidence explicit.
For cost-control topics, discuss context length, prompt tokens, completion tokens, output limits, retries, and budget protection in terms of real LLM usage.
For the comprehensive project, keep the focus on Vibe Coding with AI coding agents, real LLM integration, agent workflow, tool contracts, tests, safety, and repository-based delivery.

Verification Checklist

Before finishing a documentation change:

Run rg for stale chapter names, old paths, TODO-like editing notes, and forbidden teaching references such as 助教.
Run git diff --check.
Check README links and folder paths when renaming chapters.
If code or runnable commands changed, run the relevant smoke tests where practical.
If a command fails because of platform, dependency, model, or Docker availability, update the manual or grading document with the correct feedback path.

Git Hygiene

Do not revert user changes.
Do not stage unrelated generated outputs, caches, screenshots, model files, secrets, or local experiment artifacts.
When asked to publish, push the intended branch to the configured GitHub and Gitee remotes.
If a GitHub or Gitee account is unavailable locally, report the active remote and authentication state clearly.

MSD Course Maintenance

Purpose

Repository Shape

README.md is the course entry and topic index.

Chapter folders use sortable prefixes: 00_0, 00_1, 01 through 10, and 99.

Most chapters contain instruction.md, 实验报告提交及评分标准.md, and 附件/.

instruction.md is the experiment manual. 实验报告提交及评分标准.md is the submission and grading contract.

Keep chapter titles, README links, folder names, and grading document titles synchronized.

Teaching Assumptions

Students usually have one personal laptop.

The course has no teaching assistants and no required classroom demo session.

Documents must be self-contained enough for students to work during lab time or finish later.

Students may use AI tools for troubleshooting; unresolved problems should be discussed with the 授课老师.

Avoid references to asking assistants, receiving demo-only materials, or relying on resources outside the manual unless the user explicitly adds that teaching process.

Editing Rules

Read README.md and the target chapter before editing.

Keep edits scoped to the requested chapter, index entry, or grading document.

Keep student-facing prose in a teacher-authored Chinese style: use 同学们, 本实验, 本阶段, 授课老师.

Do not write real server addresses, accounts, tokens, API keys, internal endpoints, or private connection details into the repository.

When a manual changes required artifacts, logs, tests, model usage, environment paths, or safety requirements, update the matching grading document.

Do not put detailed submission rubrics in instruction.md when a separate grading document already carries them.

Keep chapter boundaries clear. Do not re-teach earlier chapters except as prerequisites or short reminders.

Course-Specific Priorities

Prefer Windows-laptop-friendly instructions when the audience is students; add macOS/Linux differences when they matter.

Prefer conda as the required local environment path. Docker may be recommended or used for reproducibility when feasible.

For LLM and Agent chapters, require real model evidence when the experiment says it is mandatory. Fake clients are for deterministic tests and offline troubleshooting only.

For Agent topics, make tool contracts, MCP-style interfaces, logs, failure paths, safety boundaries, and verification evidence explicit.

For cost-control topics, discuss context length, prompt tokens, completion tokens, output limits, retries, and budget protection in terms of real LLM usage.

For the comprehensive project, keep the focus on Vibe Coding with AI coding agents, real LLM integration, agent workflow, tool contracts, tests, safety, and repository-based delivery.

Verification Checklist

Before finishing a documentation change:

Run rg for stale chapter names, old paths, TODO-like editing notes, and forbidden teaching references such as 助教.

Run git diff --check.

Check README links and folder paths when renaming chapters.

If code or runnable commands changed, run the relevant smoke tests where practical.

If a command fails because of platform, dependency, model, or Docker availability, update the manual or grading document with the correct feedback path.

Git Hygiene

Do not revert user changes.

Do not stage unrelated generated outputs, caches, screenshots, model files, secrets, or local experiment artifacts.

When asked to publish, push the intended branch to the configured GitHub and Gitee remotes.

If a GitHub or Gitee account is unavailable locally, report the active remote and authentication state clearly.

msd-course-maintenance