Skip to content

skills/exam-help.md

View on GitHub
# exam-help — en student-facing pack

> Wording only. Behavior lives in [skills/exam-help/SKILL.md](../../../skills/exam-help/SKILL.md).

## Student-facing Output

### Four-step map

1. Start in lightweight on-demand processing (recommended), or explicitly choose full ingestion when a complete knowledge base is worth the extra setup.
2. `exam-tutor`: teach one current page/chapter slice at a time and persist full explanations; only full mode enters typed Guide authoring.
3. `exam-quiz`: select and grade bank questions only.
4. `exam-review` + `exam-cheatsheet`: revisit mistakes/confusions; render a printable sheet only when authorized.

### Choices and files

- Modes: from scratch, start from a chapter, or fill gaps. Time tiers: ≤1 day, 1–3, 3–7, or >7 days. Only an explicit no-questions request suppresses interaction and caps completion at `covered_unverified`.
- `processing_mode=lightweight` is the default/recommended choice: inspect only current-phase PDF pages or one definitely single-frame PNG/JPEG/BMP, with at most eight primary pages and one active batch. Contact sheets group up to four pages at roughly 768 px/tile and are overview-only. Exact external answer pages use `register-answer-dependency`; dependency pages are locator/detail context, while only source-qualified target-answer crops enter solution calls. All canonical evidence is PNG under `.lightweight/assets/`, and figure prompt/answer crops stay distinct. Only receipt-backed `abandon --reason` closes an unfinished batch; `replace-taught --reason` retains a taught predecessor/event as superseded history and plans its exact-slice successor. Keep the state machines and generate no Study Guide/PDF. Explanations remain detailed. Explicit `full` opens the complete ingestion/review route.
- `artifact_mode=chat|visual` is separate. Full processing does not silently request a PDF; only explicit full + visual enters typed Guide → render → receipt → all-page QA. A saved visual preference is dormant/effectively chat in lightweight. Never guess a subscription.
- `answer_explanation_mode=ordinary|isolated` is separate too. At full-v2 Guide entry, first perform a host-capability handshake. If official host capabilities verifiably provide a fresh independent child context and can restrict both input and tools to one exact item, the internal isolated subagent is enabled by default after one notice about extra model quota and time; no API key or external upload is involved. If any capability is incomplete (including tool restriction) or cannot be confirmed, use `ordinary` and explain why. An external Provider is a fallback only when explicitly named by the learner and still requires the two-stage billing, retention/privacy, exact-scope, current-pricing, and upload authorization.
- `preferences.interaction_style=batch|step_by_step` is an optional full-teaching cadence, not another startup choice. Stored step mode is effective only in full with `no_questions=false`; otherwise it is retained but dormant and effective cadence is batch. “Continue” is navigation, while completion needs the marked walkthrough plus its atomic `record-taught-example` binding. Older unbound teaching IDs remain valid batch history; bound IDs remain live-checked after cadence changes.
- `.lightweight/session.json` is on-demand page-batch truth; `.ingest/` is full-mode build/review truth; `study_state.json` is progress truth and `study_progress.md` its generated view; `references/quiz_bank.json` is the only scored-quiz source; `notebook/` is durable teaching.
- Ingestion exit `0` means ready or usable-with-named-gaps, `10` means content blocked, and other nonzero means an operation failure.
- Lightweight completion requires every current unsuperseded batch taught with notebook/progress bindings and can reach `covered_unverified`; `verified` requires two revision-bound checkpoints (one pass) from the unchanged quiz bank that pre-existed initialization. Startup records only an immutable stat baseline; absent-at-init, drifted, or legacy-unbound banks/checkpoints do not qualify. Routine health checks use metadata/physical identity only; exact hashes run at transitions/completion or explicit `status --verify-live`. Older taught history is `unchecked_historical` until that phase resumes.
- MinerU, Docling, and LangGraph are explicit-named-request and remote/cloud-only; never download, install, or execute them locally.

### Quiz and source card

Types: `choice`, `subjective`, `diagram`, `fill_blank`, `true_false`, `code`.

- 🟢 From your materials
- 🟡 AI-supplemented — may differ from what your teacher taught
- ⚠️ AI-generated answer — not from your teacher or textbook

Never invent a checkpoint or disguise AI content as course evidence. See [language policy](../../../docs/language-policy.md).
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall