devpilot-learn

name

devpilot-learn

description

Turn a single article, document, or web page into a standalone HTML learning artifact: either a bilingual digest or a chaptered Chinese study guide with source summaries. MUST use this skill — not raw summarization — whenever the user asks to summarize, digest, get key points from, extract takeaways from, or turn a URL, PDF, .docx, .md, .txt, paper, report, or blog post into notes or study material. Trigger on: summarize this, key points, takeaways, digest, highlights, TL;DR, what does it say, overview, 学习资料, 学习笔记, 复习资料, 总结, 摘要, 要点, 重点, 提取, 整理, 讲了什么. NOT for: writing new content, translating, comparing two documents, building apps, multi-source news, code review, or data/CSV analysis.

Learn

Generate a standalone HTML close-reading study artifact from a single source.

This skill has one job and one output shape. It does not summarize. It produces a single-column, reading-oriented document that keeps the source's actual words and helps the reader study them:

• Verbatim original — the source's substantive text, copied faithfully, is the primary prose and dominates the page.
• Chinese translation — a faithful translation sits directly below each original passage (never in a side-by-side column).
• Quizzes — a short 节后小测 after each section and a 总测 at the end, with every answer collapsed inside a <details> element.

The recurring failure mode this skill guards against is over-compression: turning the source into a short digest and losing its substance. The artifact is normally larger than the source, because it adds a translation and quizzes beneath text it has kept.

Files in this skill

File	When to load
references/output-contract.md	First — the output guarantees and hard constraints, especially the rule against over-compression.
references/workflow.md	During execution — source fetching, segmentation, generation flow, save rules, and edge cases.
references/skeleton.md	When ready to emit the HTML — the single-column verbatim-original → translation → quiz layout.
scripts/fetch_source.py	When the source is a URL — fetches the page's verbatim text. Use this instead of WebFetch, which returns a summary and silently compresses the source.
scripts/extract_pdf.py	When the source is a PDF — layout-aware verbatim extraction into real paragraphs + section headings (needs PyMuPDF).
scripts/batch_sections.py	When the source is large (≥ ~6k words or ≥ 25 sections) — groups it into ordered, word-budgeted batches so you can fan out one subagent per batch instead of emitting the whole artifact in one output-limited pass.
scripts/check_coverage.py	Before saving — the mechanical coverage gate; verifies every source section survives in the .orig blocks and names what to restore.

How to use this skill

1. Read references/output-contract.md before writing anything.
2. Follow references/workflow.md to fetch the source (verbatim, into source.txt), segment it, and generate the artifact.
3. Load references/skeleton.md for the layout when you are ready to emit the final HTML file.
4. Run scripts/check_coverage.py source.txt <artifact.html> before saving and restore anything it flags — passing the gate is required.

Keep SKILL.md as the entry point. Put detailed layout, workflow, and output spec in the reference files above rather than re-expanding them inline.

Helper dependencies (Python 3): fetch_source.py uses BeautifulSoup (bs4) and degrades to the standard library if it is absent; extract_pdf.py needs PyMuPDF (python3 -m pip install --user pymupdf) and the workflow falls back to a plain Read of the PDF if it is missing. The scripts never hard-block the skill.