SuperAPI AI Directory

wewrite-learn

Writing & research

View repository

Updated 10-15 22:37

About this skill

wewrite-learn — self-learning (edit flywheel / sample library / layout learning)

Operating conventions

  • CLI: deterministic operations go through the wewrite command (must be in PATH; if missing, guide to uv tool install wewrite, or in the repository bash install.sh)。
  • {home}: user state directory = $WEWRITE_HOME or ~/.wewrite(wewrite home can be checked). config/style/history/playbook/output/exemplars are all in {home}, not in the repository; state paths in references documents follow the same convention.
  • 读取: <路径> = actually finish reading the file with a file-reading tool before continuing, not a comment.
  • references/: this skill ships with {skill_dir}/references/; {skill_dir} inside references documents refers to this skill's directory.

Sub-function dispatch

User says Action
learn my edits / I made changes, learn from them 读取: {skill_dir}/references/learn-edits.md, execute according to its process. Supports local markdown edits and WeChat draft box sync (wewrite learn-edits --from-wechat)
learn layout / learn formatting + URL wewrite learn-theme <url> --name <name>, after extraction prompt the user to set the theme field in style.yaml
learn this article / import sample article + URL wewrite fetch-article <url> -o /tmp/article.md && wewrite exemplar /tmp/article.md -s <账号名>; default third-party
import sample article + local file wewrite exemplar <文件路径>; when the user explicitly states it is their own article, add --user-authored
view sample library wewrite exemplar --list

Purpose of the sample library: exemplars are injected into the draft prompt by framework type in the writing module (wewrite-write), used only to calibrate structure and rhythm; the viewpoints, sentences, or personal experiences in them must not be reused. After import is complete, tell the user how many articles are currently in the library and which categories and ownership tags they cover.

Value of the edit flywheel: each learning session brings the next draft closer to the user's style. In learn-edits.md, only same-scope rules that recur or are explicitly confirmed by the user become hard constraints; a single edit always remains only a soft reference.