Confusion Tracker Agent Skill

confusion-tracker — concept-confusion tracking

Purpose

Capture the learner's concept-level confusions (why / what / how-derived questions — not quiz answers) during tutoring and record them into the 「概念疑难点记录」 section of study_progress.md, building a pre-exam review list. Used by exam-tutor (while teaching) and exam-review (during the final sweep).

Activation

• During tutoring, when the learner asks a concept question matching: 「为什么…？」/「…是什么、什么意思？」/「这个公式怎么推、怎么来的？」/「…的重点是什么？」/「讲一下…」, or any clarification follow-up that is not a quiz answer.
• Skip for: pure quiz answering (right or wrong), and chit-chat that needs no concept explanation.

Inputs

• The progress-file path (e.g. study_progress.md), read at session start.
• The current chapter/phase name being taught.

Workflow

• Detect — decide whether the follow-up is a concept question (not a quiz item or its answer).
• Answer — give a concise, clear explanation grounded in the current wiki chapter. Label the source: 🟢 来自资料 for material-sourced content, 🟡 AI补充，可能与你老师讲的不完全一致 for AI-supplied background. Never present AI-added content as the teacher's.
• Record — persist the confusion: 关联章节 / 疑难点 (one line) / 解答要点 (≤2 sentences) / 状态 (default 待回顾). If study_state.json is absent and Python works, first run python "${CLAUDE_SKILL_DIR}/scripts/update_progress.py" --workspace <ws> init. The normal and ONLY state-backed write path is then python "${CLAUDE_SKILL_DIR}/scripts/update_progress.py" --workspace <ws> add-confusion --chapter <ch> --note <疑难点/解答要点> — the md table is a generated view and a hand-appended row is lost on the next render. Only when Python truly cannot run may the no-Python fallback append directly to the 「## 💡 概念疑难点记录」 table in study_progress.md, auto-incrementing the 序号 column. A nonzero state command while Python runs is a fail-loud write failure, not permission to hand-edit.
• Persist-first (notebook CLI) — the state row stays exactly as above; ADDITIONALLY persist the full explanation itself (step 2's answer, provenance labels included) so it survives outside chat: echo <explanation body> | python "${CLAUDE_SKILL_DIR}/scripts/notebook.py" --workspace <ws> add-entry --chapter <ch> --type confusion --id <slug> --title <confusion gist> (body via STDIN; same --id replaces in place; notebook/index.md rebuilds; the script resolves from the skill package root). The receipt line then carries the pack-provided link line (zh 「完整解答：notebook/chNN.md#<anchor>｜目录：notebook/index.md」, en Full explanation: notebook/chNN.md#<anchor> | Index: notebook/index.md). On a failed notebook write, TELL the student (the chat explanation already delivered stands as the copy); file-less clients keep chat-only output per exam-cram's capability dispatch.
• Confirm — tell the learner it was logged (e.g. 「已记录到疑难点」) in one short line, without breaking the teaching flow.

Output Contract

• Persist one confusion record (关联章节 / 疑难点 / 解答要点 / 状态) through update_progress.py add-confusion; initialize state first when Python works. Only a true no-Python fallback appends one row to the 「## 💡 概念疑难点记录」 table in study_progress.md (序号 auto-increments).
• Persist-first default: the full confusion explanation is ALSO written into notebook/chNN.md via the notebook CLI (--type confusion, Workflow step 3) — the state row records that the confusion exists, the notebook entry preserves the explanation itself; the receipt carries the pack-provided link line. File-less clients keep chat-only output.
• During the final sweep, read the confusion records and have the learner restate each: update 状态 in place — 待回顾 → 已回顾 when explained correctly; keep 待回顾 and re-explain otherwise. Never overwrite other skills' writes.
• Student-facing output defaults to English (Simplified Chinese if the student opened in Chinese); the persisted study_state.json.language code (zh/en/bilingual) switches it per exam-cram's dispatch rule with single-language purity.

Language packs

Student-visible wording for this skill lives in per-language packs — load the one matching study_state.json.language BEFORE emitting any student-visible output:

• 中文 → ../../locales/zh/skills/confusion-tracker.md
• English → ../../locales/en/skills/confusion-tracker.md
• 双语 → compose the zh and en packs block by block, zh first with a > EN: mirror (rules in ../../docs/language-policy.md)

Display aliases such as 中文, English, and 双语 are normalized by update_progress.py; route persisted state on zh, en, or bilingual. Unset language → the merged first-ask decides it; default English unless the student opened in Chinese.

Boundaries

• Structured progress state: when study_state.json exists it is the SINGLE SOURCE OF TRUTH — record via python "${CLAUDE_SKILL_DIR}/scripts/update_progress.py" --workspace <ws> add-confusion, update review status via set-confusion-status --id <qid>|--index <N> --status 已回顾/待回顾; never hand-patch the generated study_progress.md. If the state write fails, TELL the user; never continue as if it saved.
• Only record concept questions; never quiz or grade (that is exam-quiz).
• Concept answers carry the canonical provenance labels (🟢 来自资料 / 🟡 AI补充，可能与你老师讲的不完全一致 / ⚠️ AI生成答案，非老师/教材提供); never disguise AI-added content as teacher-provided.
• Share the progress state with exam-review: in state-backed workspaces both skills go through update_progress.py (append via add-confusion, status via set-confusion-status); only a true no-Python md-only workspace appends/updates study_progress.md in place. Never overwrite other skills' writes.