Find archived snapshots of URLs
├── http_request.py # Generic REST API client
├── pdf_extract.py # Read PDFs (optional pdf extra)
├── sql_query.py # Run read-only SQL against a local DB
├── regex_extract.py # Extract structured data from
/obsidian, /alphaxiv (helper skills)
├──▶ sources/ objective paper text extracted from pdfs
└──▶ notes/ personal notes + pdf annotations
│
▼ /collate — synthesise ingested sources + notes into a page
papers/ .md or concepts/ .md
(objective
artifacts that persist in Postgres), file generation (JSON/CSV/TXT/HTML/code as downloadable run artifacts; PDF and DOCX are deliberately off — a base64 binary does not compress in the Postgres
against where it looked like it was. A sender who attached a PDF was told Caraka could not hand a document to this agent, and the mime allowlist was four
qué hay ya en el canvas.
3. Si el usuario aporta un diagrama (PDF, imagen, sketch), descríbelo
con tus palabras antes de crear nada. Identifica:
- Cuántos routers, switches y endpoints
rows
- id: manual # short, readable, stable; the join key for citations
resource: resource_intake/ .pdf # a URL, a bundle-relative path, or a scope descriptor
title:
author: <actor, e.g. Lutron