What problem does it solve? Turning a raw PDF, article, or paper into a well-linked knowledge base entry is tedious: you must summarize it, create concept and author pages, wire up bidirectional graph edges, record citations, and keep everything deduplicated. This Skill automates that entire pipeline while keeping you in control through four review checkpoints. ## Core Features & Use Cases - Four-stage gated workflow: Draft, lint, verify, and finalize stages each pause for human review, and progress is persisted so an interrupted ingest resumes exactly where it stopped, even across sessions. - Flexible input modes: Accepts a local file path, an arxiv ID, DOI, or URL (the PDF is fetched automatically via a multi-provider resolver), or a paper title matched against a research shortlist. - Graph and citation management: Creates source pages, concept and person stubs, bidirectional graph edges, and citation links through the wiki.mjs CLI, with dedup policies that prevent duplicate pages and edges. - Use Case: You drop a new paper into raw/sources/ and say "ingest raw/sources/attention-revisited-2026.pdf". The skill drafts the source page and stubs, auto-fixes lint issues, checks the summary against the source text, and finalizes with a log entry — pausing at each gate for your approval. ## Quick Start Ask the agent to ingest a source by saying "ingest raw/sources/my-paper.pdf" or "ingest arxiv:2604.03501" and then approve or revise at each checkpoint.
