Wiki Review

Run review sessions for an Obsidian wiki that uses 内化: true and ## 内化 sections. The agent is the review teacher. The Python scripts own scanning, scheduling, state, and logs.

Use

Use this skill when you need to:

• initialize or refresh the review queue
• ask recall questions from due wiki notes
• rate a completed review with FSRS
• keep review-state.json and review-log.jsonl in sync with the wiki

Preconditions

• The project has a wiki/ directory.
• Reviewable notes live at wiki/<topic>/<article>.md.
• A reviewable note has YAML frontmatter 内化: true.
• A reviewable note has a non-empty ## 内化 section.
• Python dependencies from .agents/skills/wiki-review/requirements.txt are installed.

Workflow

1. Run sync to initialize or refresh review state.
2. Read returned issues and report them; do not invent or fill in missing user internalization content, and do not mark notes reviewable. Only make mechanical YAML repairs when explicitly asked and the fix is unambiguous.
3. Run due to collect notes that are ready for review, and read its issues and initialized output too.
4. Read the note, then ask the user to recall the idea in their own words.
5. Ask follow-up questions until the answer is clear enough to judge.
6. Choose one FSRS rating for the whole note.
7. Run submit with the rating and a short answer summary.
8. Report any due issues, and continue only with entries in the returned due list.
9. Give concise feedback and continue with the next due note.

The agent asks questions and judges recall quality. The scripts calculate schedule, store state, and write logs. Do not edit FSRS fields directly.

Commands

Run from the project root:

python .agents\skills\wiki-review\scripts\wiki_review.py --wiki-root wiki --project-root . sync
python .agents\skills\wiki-review\scripts\wiki_review.py --wiki-root wiki --project-root . due
python .agents\skills\wiki-review\scripts\wiki_review.py --wiki-root wiki --project-root . submit --article wiki/topic/article.md --rating Good --answer-summary "用户解释了核心概念。"

Useful submit ratings:

• Again: no reliable recall or a serious misunderstanding
• Hard: partial recall with substantial prompting
• Good: accurate recall of the main idea
• Easy: fluent recall and clear application

State Files

• wiki/review-state.json is the scheduling source of truth.
• wiki/review-log.jsonl is append-only review history.
• YAML review fields are display summaries for Obsidian, not the canonical scheduler state.