ingestlisted

# /ingest — read the sources Extracts the content of all (or the named) sources from `sources.yaml` into `knowledge/<source-id>/<document-slug>.md`. ## Steps 1. Read `sources.yaml`. No sources registered → point at `/sources`, done. 2. If the argument names one or more `id`s → ingest only those, otherwise all. 3. **Work incrementally**: if `knowledge/<id>/<document>.md` already exists and the source file is not newer (compare mtime), skip it. For web pages: skip if ingested less than 7 days ago (frontmatter `ingested:`), unless the user says "re-ingest". 4. Write one file per ingested document (format below). With many documents (>10), do the parallelisable extraction via an agent fan-out. 5. At the end: a summary (n new, n skipped, n failed with reason) and a pointer to the next step, `/catalog`. ## Extraction per type - **folder**: collect files recursively by `pattern` (default: `*.pdf`, `*.md`, `*.txt`, `*.html`, `*.docx`, `*.png`, `*.jpg`, `*.jpeg`). A markdown or HTML file may *link* pictures (``, `<img src=…>`): follow those links relative to the file and judge what they point at, the same way. A remote link is fetched with `figures.py fetch`. PDFs as below; DOCX → the docx skill or `textutil -convert txt` (macOS); MD/TXT taken as they are; images as under **image** below. A folder of photos or screenshots is therefore ingested without a `pattern` — say so in the summary, and ask before ingesting more than 20 i