agent-wiki-ingestlisted

# /agent-wiki-ingest The user gives you a source — a URL, pasted text, or a file path. You handle the full pipeline. ## 1. Detect the input type - **URL** — fetch the content (use web extract, agent-browser, curl, or whatever tool is available) - **Pasted text** — the user pasted content directly in the chat. Save it VERBATIM to raw/ — word for word, no summarization, no truncation. The raw file must be a complete copy of what was pasted. Not a single word lost. - **File path** — a file already on disk (may or may not be in raw/) - **Obsidian Web Clipper** — if the user clipped an article via Obsidian Web Clipper, it may already be in raw/untracked/ as markdown with images. Check for it. - **Images** — if the source contains images, save them alongside the markdown in raw/. Reference them in the source page. Images give context that text alone misses. ## 2. Save to raw/untracked/ Save the source as a markdown file in raw/untracked/ with a dated name: `raw/untracked/YYYY-MM-DD-short-slug.md` If the source is already in raw/, skip this step. ## 3. Read, analyze, and discuss - Read the full source content - Read the relevant existing wiki pages - Summarize the key takeaways (3-5 bullet points) Then ask 2-5 socratic questions based on what the source says vs what the wiki knows: **Project:** "This looks related to [project]. Correct?" (don't ask "which project?" if you can guess) **Contradictions:** "The source says [X] but the wiki says [Y]. Which is current?" **Deci