AGENT SKILL

study

Daily review session — show due cards, collect ratings, advance the schedule

License 许可
Apache-2.0
所属仓库
boltguo/aouo
最近更新
2026-11-12
Stars
★ 2
Forks
0

01 前言 Front Matter

name
study
pack
vocab
command
true
triggers
[ "study", "review words", "vocab session", "背单词", "复习" ]
description
Daily review session — show due cards, collect ratings, advance the schedule

03 SKILL.md 正文

Study

Walk the user through today's due cards (and a small handful of new ones), collect Again/Hard/Good/Easy ratings, advance the schedule.

Goal

Complete one focused session: up to daily_target cards (default 15). Log every rating into reviews. Update each card's interval_index and next_due.

Procedure

  1. Load the queue Use the db tool to pull:

    • Due cards: SELECT * FROM cards WHERE next_due <= date('now') AND status != 'mastered' ORDER BY next_due, id LIMIT ? (limit = daily_target).
    • If the due pool is empty, top up with new cards: WHERE status = 'new' LIMIT ?.
  2. Empty queue

    • If both are empty, say: "Nothing due today. Want me to generate 10 new cards at your level?" — only generate on user confirmation.
  3. Quiz loop (per card)

    • Show the front: word + ipa (no definition).
    • Ask the user to recall (free-text or "show").
    • On "show" (or after their guess), reveal def_zh, def_en, example_en.
    • Ask for a rating with msg type: "keyboard": Again / Hard / Good / Easy.
  4. Advance the schedule Read the active interval curve once at session start:

    • SELECT days_csv FROM intervals WHERE id = (SELECT v FROM user_state WHERE k = 'active_interval_id').
    • Parse days_csv into an array D = [d0, d1, ...].

    For each rating:

    • Again: interval_index = 0, lapses += 1, next_due = date('now', '+1 day'), status = 'learning'.
    • Hard: keep interval_index (re-show at same step), next_due = date('now', '+' || D[interval_index] || ' days').
    • Good: interval_index = min(interval_index + 1, len(D) - 1), next_due = date('now', '+' || D[interval_index] || ' days'), status = 'review'.
    • Easy: interval_index = min(interval_index + 2, len(D) - 1), same next_due math. If at the last step, mark status = 'mastered'.

    Log the rating:

    {
      "action": "practice",
      "skill_type": "vocab.study.review",
      "content": "<word>",
      "metadata": "{\"card_id\":42,\"result\":\"good\",\"response_ms\":3200}"
    }
    
  5. End-of-session summary

    • "Reviewed 12 cards. 8 good, 2 hard, 2 again. Next session: tomorrow 08:00."
    • Increment user_state.total_reviewed.

Rules

  • Never skip the rating step — even a quick "good" is needed to advance.
  • One card per turn; let the user breathe.
  • If the user types "stop" or "later", end gracefully and update the session count even if partial.
  • Don't reveal the def_zh / def_en before they make at least one guess (or explicitly say "show").

04 同仓库更多 Skills