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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/Morrison-Lab/ai-config --skill plan-review-session命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | plan-review-session |
| description | Turn student errors into lessons. |
| user-invocable | true |
| allowed-tools | ["Bash","Read","Edit","Write"] |
Convert a ranked list of common mistakes (e.g. from grade-work) into a review session: a self-contained chapter that, for each common error, pairs a "common mistake" callout with a worked correction, ordered most-impactful first.
Origin: built from a midterm review chapter that distilled grading findings
into a new chapters/<name>.qmd in a Quarto course book.
Per error category, in this order:
## heading that states the correct principle (not the
mistake) — e.g. "Nelson-Aalen estimates the cumulative hazard first, then
exponentiates".::: {.callout-warning} block with a ## Common mistake sub-heading
describing what students did wrong (anonymized — patterns, never names).::: {.solution} block working through the correct approach with concrete
numbers, reproducible R where it helps.Order categories by frequency × consequence; lead each part with its most
consequential error. Close with a summary table cross-linking each mistake to
its section (@sec-...).
Match the existing chapters — read one first before writing.
title: and the html/revealjs/pdf/docx
format block, then the book's shared-config include (which pulls in the R
setup and any LaTeX macros).::: {#tbl-...} / ::: {#fig-...} with the
caption as the last line inside the div — not chunk-option tbl-cap.#| code-fold: true for figure/table chunks; keep R
simple and runnable.- chapters/<name>.qmd to the book's _quarto.yml under the right
part. (Some books split the config across separate book and website YAMLs —
if so, add the chapter to each project: render: list and to the
website: navbar: menu.)Run all three and fix what your change caused:
quarto render chapters/<name>.qmd --to html # no errors/warnings
Rscript -e 'lintr::lint("chapters/<name>.qmd")' # no lints in your code
Rscript -e 'res <- spelling::spell_check_files("chapters/<name>.qmd", ignore = readLines("inst/WORDLIST")); print(res)'
?@) and that tables/computed values resolved.spell_check_files on .qmd does not strip
in-math content, so it false-flags LaTeX macro tokens (mathbf, tfrac,
ge, …). CI (clean container) does not flag these. So only add genuine
prose words to inst/WORDLIST (e.g. exponentiate); do not pollute it with
math macros.origin/main (git checkout -b <name> origin/main); never commit
to main directly.git add -A.gh pr create, request a reviewer (see request-pr-review).