Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add ZeKaiNie/universal-examprep-skill --skill exam-helpgit clone --depth 1 https://github.com/ZeKaiNie/universal-examprep-skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/zekainie/universal-examprep-skill/exam-help)<a href="https://agentmods.dev/skills/zekainie/universal-examprep-skill/exam-help"><img src="https://agentmods.dev/badge/skills/zekainie/universal-examprep-skill/exam-help/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/zekainie/universal-examprep-skill/exam-help"><img src="https://agentmods.dev/badge/skills/zekainie/universal-examprep-skill/exam-help.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00062 | $0.01180 |
| Opus 5 | $0.00031 | $0.00590 |
| Sonnet 5 | $0.00012 | $0.00236 |
| Haiku 4.5 | $0.00006 | $0.00118 |
Grade A, and why
exam-help scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.
exam-help — quick-reference card
Purpose
Render one read-only card covering the validated workflow, three modes, four time tiers, lightweight|full, chat|visual, workspace truth/views, six quiz types, provenance, and subskill routing.
Activation
Use only when the student asks how the suite works, which modes/types exist, or what workspace files mean.
Inputs
No files, arguments, or state reads. The caller supplies the selected language.
Workflow
- Emit the matching static language-pack card: persisted
zh,en, orbilingual, otherwise an explicit ad-hoc request. - Read no wiki, bank, plan, progress, or workspace file; run no script/subskill.
- Stop without tutoring, quizzing, ingesting, grading, or initialization.
The card must say:
processing_mode=lightweightis the default/recommended startup choice. It inventories names, then visually processes only the current-phase PDF pages or definitely single-frame PNG/JPEG/BMP (maximum eight primary pages and one active batch). Overview contact sheets group at most four pages at roughly 768 px/tile; page/prompt/dependency detail and target-answer-only solution calls use source-qualified locations and canonical PNG evidence under.lightweight/assets/. Bind exact external answer pages withregister-answer-dependency; dependency pages are locator/detail only. Figure prompt/answer crops stay distinct. An unfinished batch can close only through receipt-backedabandon --reason; taught progress cannot be abandoned, whilereplace-taught --reasonpreserves it as superseded history and plans an exact-slice successor. It keeps the page-batch / progress state machines and creates no Study Guide/PDF.processing_mode=fullis explicit opt-in and opens the complete ingestion/review route. Input-token savings never shorten the teaching explanation.artifact_modeis independent from processing intensity and never inferred from subscription. Missing/legacy/unknown meanschat. In full mode,chatstops without PDF while standingvisualadditionally requires typed Guide, render, receipt hashes, every-page QA, andartifact_ready=ready. A one-shot artifact leaves stored state unchanged. In lightweight, a savedvisualpreference is dormant and effective output remainschat; it never builds a Study Guide.answer_explanation_modeis also independent. Its stored-schema fallback isordinary, and every full-Guide item still gets a detailed beginner-first explanation. At full-v2 Guide entry, a verified host-native child with a fresh independent context plus exact single-item input/tool restrictions makesisolatedthe default unless the learner opted out; tell the learner once about extra host quota/time, with no second API key or external upload. Missing or incomplete capability staysordinaryand is stated honestly. A separately billed external Provider is explicit-request-only and retains no-upload planning, current pricing/privacy disclosure, and exact-plan upload consent. A model name, subscription, key,full, orvisualalone proves neither route.ingest_course.pyis the full-mode build entry; exit 10 is process success with blocked readiness, and teaching remains forbidden..ingest/is full-mode build/review truth,.lightweight/session.jsonis on-demand page-batch truth,study_state.jsonis progress truth, and Markdown is a generated view.- Lightweight completion uses current-phase taught-batch/notebook events and skips
typed Guide/full-build evidence; superseded predecessors/events remain history but
leave the current denominator.
verifiedneeds revision-bound checkpoints from an unchanged bank that pre-existed lightweight initialization; startup captures only its immutable stat baseline. Routine health checks use metadata/physical identity, not stream hashes; exact hashes occur at transitions/completion or explicitstatus --verify-live. Older taught history isunchecked_historicaluntil that phase is resumed. - MinerU, Docling, and LangGraph are named-request-only and remote/cloud-only. Never download, install, probe, import, or execute them locally.
- Provenance is exact: 🟢 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.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 92 lines · 62 tokens per session scan A 1b20be421ff9
exam-help is a skill published in the GitHub repository ZeKaiNie/universal-examprep-skill (281 stars, last pushed 7d ago), licensed MIT. It adds 62 tokens to every session and 1,180 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
bq-skill
A coaching workflow for building a reusable library of truthful stories for behavioral job interviews. It uses structures such as STAR—Situation, Task, Action, Result—and can connect stories to a job description and résumé.
PersonaLingo
A guided tool for creating personalised IELTS speaking practice materials. IELTS is an English exam, and the tool turns a learner’s answers and conversation into a personal profile and a static study website.
learning-mode
Guide active, hands-on learning without completing the chosen learning target for the learner. Use when the user explicitly wants to learn, practice, master, or build something themselves with guidance, especially when a technology or concept is new to them. Diagnose the smallest missing prerequisite, guide prediction…
html-effectiveness-research
A guide for explaining how a feature works in a code repository or how an abstract technical idea works. It can structure the explanation with a summary, expandable steps, code views, questions and answers, or an interactive visual.
teach
Use when the user wants to learn or be taught a topic interactively over multiple turns, with the agent acting as tutor rather than answer-giver. Triggers on phrases like "teach me", "tutor me", "quiz me", "let's learn", "help me understand X by working through it step by step", or any request framed around learning a…
rwkv-architecture
RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.