一键导入
pivot
Use when pivoting a finalized learning sheet's task and examples to a new real-world context, or when scrutinizing its reasoning for logical gaps
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。
菜单
Use when pivoting a finalized learning sheet's task and examples to a new real-world context, or when scrutinizing its reasoning for logical gaps
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
Use when generating a complete week of materials, including learning sheets and test bundles, through adversarial writer-vs-reviewer debate
Use when an instructor wants to wire up Zulip-based homework submission for the course — creates a grading bot on the instructor's Zulip realm via API, writes `.zuliprc` at repo root, and bootstraps `coursedesign/roster.csv`. Run once after `/bootstrap`. Triggers on phrases like "set up zulip submissions", "configure zulip grading bot", "wire up the homework bot", "/setup-zulip-grading".
Use when writing a new learning sheet from scratch for a specific week — reads textbook sections, follows templates, and produces a single learning-sheet.typ file
Use when an instructor wants to pull homework photos that students DM'd to the grading bot, grade them, and DM per-question feedback back. End-to-end Zulip-based grading orchestrator that delegates to /grade-homework and (optionally) /homework-report. Requires /setup-zulip-grading to have been run once. Triggers on phrases like "grade homework via zulip", "pull and mark week N from zulip", "/zulip-mark-feedback 3", "send the feedback DMs".
Use when setting up a new course from a forked template repo — gates on a canonical source (textbook or equivalent), ingests its title-page + ToC for metadata, optionally folds in supplementary syllabus/slides/notes, writes config.toml, extracts textbook chapters, creates the weekly schedule, and enables GitHub Pages deployment when possible
Use when the user wants to grade a folder of student homework submissions against a teacher-provided solutions document — handles mixed formats (PDF, JPG, DOCX), produces a grades CSV and per-student English feedback with explicit flagging of ambiguous items for teacher review. Triggers on phrases like "grade the homework", "mark HW9", "批作业".
| name | pivot |
| description | Use when pivoting a finalized learning sheet's task and examples to a new real-world context, or when scrutinizing its reasoning for logical gaps |
| argument-hint | [week-number] |
Week number is provided as $ARGUMENTS. Replace N below with the week number.
Read these files before starting:
weekN/N.learning-sheet.typ — the finalized sheet to pivottextbook/*.md chapters (check coursedesign/schedule.typ for assigned textbook sections)week1/1.learning-sheet.typ — reference for how task boxes and examples are structured (skip if pivoting week 1 itself; use templates/week-template/N.learning-sheet.typ instead)Extract from the current sheet:
#badge-broader and #badge-deeper) that reference the task contextGenerate candidate motivating tasks for the same theory. Each must be:
Brainstorming-style presentation: Before presenting options, write a short analysis of the current task's strengths and weaknesses (e.g., which of the week's big ideas it covers vs. misses). Then present candidates using AskUserQuestion with markdown previews. Each option must include:
description fieldInclude the current task as one option (in case the user wants to keep it and only do the reasoning audit).
Walk through each cataloged item from Step 1 one by one. For each item:
AskUserQuestion to let the user approve, request changes, or skipProcess items in this order:
Only apply each edit after the user approves it. If the user requests changes, revise and re-present before moving on.
Do NOT change:
Independently from the pivot, scrutinize every non-trivial reasoning chain in the sheet. This includes formal proofs, but also informal arguments like "a stack reverses, so PDAs can't do X" or "this is equivalent to Y because...".
For each argument, ask:
## Reasoning Audit: Week N
### Argument 1: [name] (lines NN-NN)
- **Claim:** [what is being argued]
- **Status:** sound / has gap / uncertain
- **Gap (if any):** [what's missing or hand-wavy]
- **Suggestion:** [how to fix or strengthen]
### Argument 2: ...
(repeat for each non-trivial argument)
### Summary
- X/Y arguments are sound
- Gaps found: [list]
List non-sound items — collect all arguments with status "has gap" or "uncertain" into a numbered checklist for the user.
Confirm each item one by one — for each non-sound item, use AskUserQuestion with markdown previews to present:
markdown field; "Skip (no change)" option shows the current text in its markdown fielddescriptionOnly apply fixes the user approves. Process items sequentially (one AskUserQuestion per item).
.typ file + the reasoning audit report