菜单
Review the user's solution for one studyground exercise. Reads exercises/<name>/main.py, gives feedback, writes it to exercises/<name>/feedback.md and to a feedback block in the calling lesson.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Answer an inline question marker (?> or ?>>) in a lesson file. Locates the Nth marker of the given kind and writes the answer in-place. Use when /api/ask routes a user question.
Have a real first conversation with a new learner — open-ended, learner-led, not a survey. Let them tell you what they're after; follow up where it actually matters; when you have enough to draft something useful, offer to write the curriculum. Then on action=finalize, write tracks/<track>/curriculum.md.
Generate the next lesson in the current track. Reads progress.json, writes a new lesson file, updates progress. Use when the user is ready to advance.
Compact a lesson by folding answered ?> Q&A pairs into <details> blocks, so a re-read flows past them. Run after the user has worked through a lesson.
A persistent tutor / coordinator that the learner talks to ABOUT a course (not inside a single lesson). Knows the curriculum, what's done, what's pending, the materials, and recent btw threads. Discusses plan, pace, gaps, suggestions — does not write to lesson files itself; recommends actions the user can take with other skills.
Start a new learning track on a topic. Generates the first lesson under $STUDYGROUND_DIR/lessons/ and updates progress.json. Use when the user says "let's learn X" or invokes /studyground:learn.
| name | check |
| description | Review the user's solution for one studyground exercise. Reads exercises/<name>/main.py, gives feedback, writes it to exercises/<name>/feedback.md and to a feedback block in the calling lesson. |
Review the user's attempt at one exercise.
You are called by /api/check. The request gives:
lesson — slug like 01-attention (where the :::exercise block was authored)exercise — name like linear-attention (corresponds to exercises/<name>/)Read exercises/<exercise>/main.py (and any other .py files there). If main.py is unmodified from the scaffold (# TODO: implement still present and no other code), say so politely and stop — don't fabricate feedback for empty code.
Read exercises/<exercise>/README.md to recall the goal.
Read the source :::exercise <exercise> block in lessons/<lesson>.md for the original framing.
Review the code:
Write your review to exercises/<exercise>/feedback.md, replacing any existing content. Use markdown. 2–6 paragraphs. Lead with the bottom line ("Looks correct, here's one concern…" or "Close, but the shape is off in the matmul step…"). Don't open with "Great work!" / "Excellent job!" or other warm-up boilerplate — write like a code reviewer, not a cheerleader.
Append (or replace if it exists) a feedback block in lessons/<lesson>.md, immediately below the matching :::exercise <exercise> ... ::: block. Use this exact format:
<!-- feedback:start name="<exercise>" -->
*Last checked: <YYYY-MM-DD>* — one-sentence summary.
<details>
<summary>Full review</summary>
{full review markdown here}
</details>
<!-- feedback:end -->
If a <!-- feedback:start name="<exercise>" --> ... <!-- feedback:end --> block already exists for this exercise, replace it. Don't accumulate stale reviews.
main.py or test_main.py — those are the user's workspace.progress.json from this skill.main.py has obvious off-the-rails issues (wrong language, totally unrelated code), say so plainly — don't pretend it's "almost there".test_main.py with pytest and include the result in your feedback (PASS/FAIL summary at top, then a quick interpretation if anything failed). Otherwise the review is static-only.