learn-applylisted

Study any source (web URL / GitHub repo / video / podcast / pasted text or local file), extract the key learnings, then map them onto a specific project — or, if none applies, onto a concrete real-life problem — with worked examples. Invoke with /learn-apply or /lp, or when the user says "lp", "learn-apply", or pastes a source and asks how it helps their work. Do NOT auto-trigger on a plain "summarize this" — only when the user wants the applied learnings → project / real-life mapping.
prodikeycw/learn-apply-skill · ★ 0 · Data & Documents · score 72

Install: claude install-skill prodikeycw/learn-apply-skill

# learn-apply (alias: lp) Turn any input source into **applied** learnings for a real project — not just a summary. ## Step 1 — Detect the source type and read it Route by what the user gave you. Prefer tools already installed on the machine; introduce no new dependencies. | Source | How to read it | |---|---| | Web page / article URL | `markitdown "<url>"` (install with `pipx install markitdown`, using Python 3.12 — 3.14 installs a broken build) | | GitHub repo or file | Use `gh`, never a raw web fetch (repos may be private): `gh repo view OWNER/REPO`, README via `gh api repos/OWNER/REPO/readme --jq .content \| base64 -d`, files via `gh api repos/OWNER/REPO/contents/PATH --jq .content \| base64 -d` | | Video / podcast (YouTube, etc.) | Transcribe with whatever tool you have (e.g. `yt-dlp` + a Whisper CLI), then read the generated transcript. If it needs login cookies and fails, surface the error, don't silently retry | | Pasted text | Read directly | | Local file (.pdf/.docx/.pptx/…) | `markitdown "<path>"` | | Image (screenshot / photo / scanned page) | **Read it directly** — Claude Code / Codex render images visually, so you read the text (incl. tables, captions, CJK) and layout without a separate OCR step. Plain `markitdown` does NOT OCR images; `tesseract` is an option only if installed. Flag low-confidence for blurry or handwritten scans. | | URL blocked (403 / paywall) | Fall back to your client's web-fetch tool, then a web search across secondary sources — and **f