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 AlexBybye/SCUT_CS --skill material_convertergit clone --depth 1 https://github.com/AlexBybye/SCUT_CSWrote 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/alexbybye/scut_cs/material_converter)<a href="https://agentmods.dev/skills/alexbybye/scut_cs/material_converter"><img src="https://agentmods.dev/badge/skills/alexbybye/scut_cs/material_converter/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/alexbybye/scut_cs/material_converter"><img src="https://agentmods.dev/badge/skills/alexbybye/scut_cs/material_converter.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 | $0.00000 | $0.01339 |
| Opus 5.5 | $0.00000 | $0.00536 |
| Sonnet 5.5 | $0.00000 | $0.00268 |
| Haiku 4.5 | $0.00000 | $0.00134 |
Grade A, and why
material_converter 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 1mo 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 — 73 lines — stays where its author put it; the contents beside it link to each section on GitHub.
技能:学科资料转 Markdown(material-to-markdown)
收录于仓库
apps/tools/material_converter/SKILL.md。此技能描述“如何用 AI 把学科资料/下的文件转成合格的知识库 Markdown”,并指明 AI 能做什么、绝不能做什么。
能做什么
- 按
apps/scut-senior/docs/MATERIAL_TO_MARKDOWN_SOP.mdv1.7 走完整转换流程; - 调用确定性抽取工具
apps/tools/material_converter拿到忠实骨架(标题/段落/表格/图片、 原生 OMML→LaTeX、page/slide/heading 锚点、manifest 记录); - AI 语义归一化(工具不含、需要由 AI/模型完成):
- 扫描稿/手写稿的 OCR 校正、版面与阅读顺序恢复;
- 把 MathType/OLE 公式预览图转成 LaTeX(只转能逐项对得上原件的);
- 给出历年题的题目边界候选,供人工确认;
- 对低置信 OCR 位置做重点检查并写
notes。
不做什么(SOP §4 红线)
- 不总结、缩写、解释、纠错、补写、或根据常识/答案猜公式、数字、单位、术语、代码;
- 不把工具或 AI 输出直接标
passed;最终结论由人工审核决定; - 不改动原文;原资料疑似有错时原样保留,在
notes说明; - 不破解加密 zip(准入判断跳过),不上传第三方云解析,不执行宏。
执行流程
- 课程接入:先确认
packages/contracts/v1/courses.json已注册该课程(course_id/aliases/repository_paths)。 - 隐私前置:含学生/贡献者姓名、班级、学号的文件名先
git mv脱敏;元数据/正文真实 身份信息清除或记notes;试卷密封线模板字段保留。 - 确定性抽取:
cd apps/tools/material_converter .venv/bin/python -m material_converter.main --course <文件夹名> --validate - AI 语义归一化:
无视觉模型时跳过 4b/4c,直接人工回填 formulas.json 后执行 4d。# 4a. 导出作业包(公式图清单 + OCR 页清单) .venv/bin/python -m material_converter.main --emit-ai-jobs # 4b. GLM-4V 视觉转写公式图(三道闸:三票多数决 → 确定性校验 → mathtext 渲染闸; # 凭证放仓库根 .cache/glm4v.env,见 README;未过闸的自动保留 PNG) .venv/bin/python -m material_converter.main --vision-run --vision-workers 4 # 先试小样: --vision-run 20 # 4c. 按内容哈希传播转写结果到全部作业包(同一张图只转一次) .venv/bin/python -m material_converter.main --vision-propagate # 4d. 应用回知识库(替换为 $...$、清理已用资产、状态保持 pending) .venv/bin/python -m material_converter.main --finalize - 人工审核:逐文件对照原件,确认公式/题界/顺序/隐私,然后才把该行置
passed。
关键约定
manifest.title是source_title的事实源;course/title/frontmatter 必须一致;- 每份资料归一门课;同资料误放其他目录只处理一份规范来源;无答案版是答案版严格子集时只转答案版;
- 无文本层/乱码 PDF 采用整页图片方案(本仓库既定模式),
notes注明“待人工决定 OCR”; - 转换中间产物(
.work/、.ai_jobs/、staging、调试 HTML)不入 Git;knowledge/不留临时产物。
六条红线(每条都有真实事故背书,跑批前必读)
- 修复必须进管线代码+回归用例;禁止对生成物做一次性脚本后处理(复发过)。
- soffice 等模块级配置在使用点惰性解析;每个调用点自检(曾静默 kept-as-vector=1024)。
- 从 manifest 的 course 拼磁盘路径一律经
knowledge_dir()(probability 是 legacy 目录, 直接拼 course_id 会静默扑空、误删文件)。 - 过滤条件先打印 distinct 值核对再执行;删除类操作先输出将删清单(字段值写错过三次)。
- 清→跑→验证压缩进单次调用;破坏性操作前先 tar 快照(备份两次挽救误删)。
- 题目锚点:试卷类用强+弱信号(中文序号/第X题/阿拉伯顿号 + 数字点号/括号号,排除小数), 练习解答类只用强信号防列表误标;锚点是工具提议,须人工确认。
What ships with it
19 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- bootstrap.ps1 1.6 KB runs code
- bootstrap.sh 1.4 KB runs code
- material_converter/__init__.py 0 B runs code
- material_converter/__main__.py 48 B runs code
- material_converter/ai_stage.py 7.2 KB runs code
- material_converter/convert.py 24 KB runs code
- material_converter/courses.py 3.1 KB runs code
- material_converter/docx2md.py 22 KB runs code
- material_converter/main.py 13 KB runs code
- material_converter/omml2latex.py 15 KB runs code
- material_converter/pdf2md.py 2.9 KB runs code
- material_converter/ppt_legacy.py 1.9 KB runs code
- material_converter/pptx2md.py 4.2 KB runs code
- material_converter/propagate_vision.py 2.5 KB runs code
- material_converter/vision_worker.py 10 KB runs code
- probe_report.py 3.7 KB runs code
- README.md 9.2 KB
- requirements.txt 104 B
- spot_check_report.py 7.7 KB runs code
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.
- 1mo ago First seen · 73 lines · 0 tokens per session scan A 22aebc8f3f07
material_converter is a skill published in the GitHub repository AlexBybye/SCUT_CS (129 stars, last pushed 1mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,339 tokens. 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
lineage-skill
Turn courses, books, video, audio, PDFs, slides, transcripts, OCR, notes, and long-form materials into source-grounded learning Skills that teach unseen concepts progressively with reliable terminal ASCII or SVG visuals, ask two end-of-lesson questions together by default, give focused feedback, schedule review, test…
tutor-setup
Transforms knowledge sources into an Obsidian StudyVault. Two modes: (1) Document Mode — PDF/text/web sources → study notes with practice questions. (2) Codebase Mode — source code project → onboarding vault for new developers. Mode is auto-detected based on project markers in CWD.
deep-research
Deep research report generation for Reviva's built-in deep-researcher agent. Use for learning and education research, teacher preparation research, document-based synthesis, literature/source reviews, office research, industry analysis, competitor scans, policy interpretation, and any request for a cited long-form…
feynman-learning-coach
Guide users through the Feynman Technique by asking them to explain a concept in simple language, diagnosing gaps, correcting misunderstandings, simplifying explanations, and retesting understanding. Use for deep learning, concept mastery, document-based study, exam preparation, and teach-back practice.
socratic-tutor
Guide users through learning by asking questions, giving progressive hints, and helping them reason step by step instead of immediately giving answers. Use for homework help, problem solving, concept learning, exam preparation, programming practice, and document-based learning.
lab-report-writer
Use for writing, editing, polishing, or structuring lab reports and course experiment reports across disciplines. Handles cases with a provided template, raw data, images, notes, or only a brief requirement, and can produce DOCX or Markdown deliverables.