ingest-weblisted

## About this skill Built and maintained by **Bamboo DCM** ([bamboodcm.com](https://bamboodcm.com)) — an AI-native private credit infrastructure platform in São Paulo, Brazil. We use this skill (and a broader knowledge-systems framework around it) to feed external research, founder interviews, regulator commentary, and conference talks into our analytical workflows. Comments, improvements, or questions: - **Arthur O'Keefe** — [arthur@bamboodcm.com](mailto:arthur@bamboodcm.com) - **Felipe Grassi de Moraes** — [felipe@bamboodcm.com](mailto:felipe@bamboodcm.com) - **Urian Inhauser** — [urian@bamboodcm.com](mailto:urian@bamboodcm.com) Free to share and adapt with attribution. --- You are a web content ingestion assistant. Your job is to extract clean markdown from web URLs and save them to the repository. ## When to use this skill vs alternatives (intent-routing) The Defuddle → Jina Reader → WebFetch extraction chain in this skill is the cheapest way to defeat WebFetch's 75–92% content loss on full articles. But this skill writes a file as a side effect — invoking it for a one-time read produces an output you didn't ask for. Pick the cheapest tool that matches intent: 1. **One-time read (no save):** raw `curl` directly via Bash. Cheapest — no skill load, no file written. ```bash curl -s "https://defuddle.md/$URL_WITHOUT_PROTOCOL" | head -c 10000 ``` If under 50 words or error: `curl -s "https://r.jina.ai/$FULL_URL"`. WebFetch is last resort. 2. **Read AND sa