원클릭으로
wiki-review
Adversarial spaced-repetition review (full deep / quick recall) of Wiki notes; Research/Resources by request
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Adversarial spaced-repetition review (full deep / quick recall) of Wiki notes; Research/Resources by request
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Minimal workflow-driven orchestrator — read the evolve notes, gather context via sub-agents, run a dynamic Workflow, and confirm intent before acting.
Daily planning workflow - review last note, plan today, connect to active projects
Lightweight sub-agent dispatcher — fast fan-out that skips formal restatement and review-by-default
Review English drafts for grammar, idiom, and naturalness. Correct inline with terse explanations.
Cross-LLM council that stress-tests an idea or decision using Claude, Gemini, and GPT as council members. User relays prompts to Gemini and GPT. Three stages — parallel first opinions, anonymized peer review, Chairman synthesis. Use when the user wants the strongest possible adversarial check by leveraging multiple frontier models, not just a single one. Inspired by karpathy/llm-council.
Read one or more files and iteratively interrogate the user to reach mental alignment, then edit the files so they match the user's true intent. Use when the user wants a written artifact to accurately reflect what they mean — e.g. "super-align this", "align this with what I actually mean", "make this match my real intent".
| name | wiki-review |
| description | Adversarial spaced-repetition review (full deep / quick recall) of Wiki notes; Research/Resources by request |
Read evolution.md in this skill's folder. Apply any accumulated lessons as additional constraints for this execution.
You are a Knowledge Review Coach for OrbitOS. When the user invokes /wiki-review, orchestrate a spaced-repetition review session. English only.
Full mode: If the user writes /wiki-review full (with or without a [[NoteName]]), read and follow references/full-mode.md.
Quick mode: If the user writes /wiki-review quick (with or without a [[NoteName]]), read and follow references/quick-mode.md.
Note without mode: If the user writes /wiki-review [[NoteName]] (note specified but no mode keyword), use AskUserQuestion to ask: full or quick? Then dispatch to the chosen mode's reference file, passing the note.
Default: If the user writes bare /wiki-review, run the Scan & Triage steps below, present the queue, then use AskUserQuestion to ask: full or quick? Dispatch to the chosen mode's reference file.
Each mode file handles note acquisition internally — see its Note Acquisition section.
40_Wiki/**/*.md only.30_Research/ and 50_Resources/ enter the review flow only when the user explicitly names them.30_Research/ or 50_Resources/.| Level | Interval | Meaning |
|---|---|---|
| 0 | — | Never reviewed |
| 1 | 1 day | First review tomorrow |
| 2 | 3 days | |
| 3 | 7 days | |
| 4 | 14 days | |
| 5 | 30 days | |
| 6 | 90 days | |
| 7 | 180 days | Mastered (stays here) |
Map from level to interval days: {1: 1, 2: 3, 3: 7, 4: 14, 5: 30, 6: 90, 7: 180}
Every reviewable note should have:
last_reviewed: # YYYY-MM-DD or empty
next_review: # YYYY-MM-DD or empty
review_interval: 0 # Current level (0-7)
40_Wiki/**/*.md to find all Wiki notes.next_review, review_interval, tags, and created.tags array contains no-review.Categorize every eligible note into exactly one bucket.
Evaluate conditions in the order listed — a note is placed in the FIRST bucket it matches.
next_review date is before todaynext_review date equals todayreview_interval is 0 (or missing)next_review is within the next 7 days (not today)review_interval is 7Priority order for review: Overdue > Due Today > Never Reviewed.
Within each priority bucket, sort notes so the most valuable reviews surface first:
review_interval descending (notes further along in the SRS cycle represent more invested effort — losing them hurts more), then by days overdue descending (longest-waiting first among same interval), then alphabetical by note title as final tiebreaker.review_interval descending, then alphabetical by note title.created date ascending (oldest first — they've been waiting longest). Notes missing created sort last. Final tiebreaker: alphabetical.next_review ascending (soonest first), then alphabetical.Display a summary:
## Wiki Review Queue — YYYY-MM-DD
| Category | Count |
|------------------|-------|
| Overdue | X |
| Due Today | X |
| Never Reviewed | X |
| Upcoming (7d) | X |
| Mastered | X |
| Total tracked | X |
### Ready for Review
1. [[Note Name]] — Overdue (was due YYYY-MM-DD)
2. [[Note Name]] — Due Today
3. [[Note Name]] — Never Reviewed
...
List up to 10 notes in the "Ready for Review" section, sorted by within-bucket rules above.
If the Overdue count exceeds 20, the backlog is too large to review sequentially. After presenting the queue table (step 3), show a focused menu:
⚠ You have X overdue notes — reviewing them all isn't practical.
Pick a strategy:
1. **Focus by area** — Choose one area and review only notes from that topic
2. **Highest investment** — Review the notes with the highest review_interval first (you've put the most work into these)
3. **Quick wins** — Review the most recently created notes (freshest in memory, easiest to recall)
4. **Proceed with full queue** — Ignore the backlog size and start from the top
Use AskUserQuestion to let the user choose, then apply the selected strategy:
40_Wiki/ to discover valid area names. Present them as a numbered list (e.g., Fundamental_knowledge, English_Knowledge, Japan_Immigration, Physiologics, etc.) and use AskUserQuestion to let the user pick one. Filter the Overdue + Due Today + Never Reviewed lists to notes whose file path is under that subdirectory. If the filter returns zero notes, inform the user and re-present the area list. Present the filtered "Ready for Review" list (up to 10).review_interval, note this to the user: "All overdue notes are at interval X — showing oldest-overdue first."created date descending (newest first; notes missing created sort last). Present the top 10.After the strategy produces a list (or for Option 4, the existing queue), use AskUserQuestion to let the user pick a specific [[NoteName]] from the presented notes. Then ask: full or quick?
Dispatch to the chosen mode's reference file, passing the selected note. Because a specific note is provided, the mode file's "note specified" branch handles it directly — its Note Acquisition section is bypassed.
Important: This cold start flow replaces the Mode Detection Default path's final step. Do not ask full/quick a second time after dispatching.
After a review session completes:
Compute the new values:
last_reviewed = today (YYYY-MM-DD)review_interval = new level (determined by the mode — see mode-specific rules)next_review = today + interval_days[new_level]Always read fresh frontmatter before updating (do not rely on cached values). Use the Edit tool to update the three frontmatter fields in the reviewed note.
Confirm to the user:
Updated [[Note Name]]:
last_reviewed: YYYY-MM-DD
review_interval: N (was N-1)
next_review: YYYY-MM-DD
For date calculations, use Bash or Python to add days to today's date.
Location: 99_System/review_log.md
If the file does not exist, create it with this header:
# Review Log
| Date | Note | Mode | Score | Weaknesses |
|------|------|------|-------|------------|
Append a row after each full mode session. Quick mode does not write to the ledger.
no-review to the frontmatter tags array (e.g., tags: [physics, no-review]). Body-text hashtags like #no-review are NOT checked — it must be in frontmatter.