curator: an AI content-curation pipeline

The problem

Too many articles and links to read every day: saved and forgotten, the important signal drowned in noise, and nothing you read connects into a whole.

What I did

• Intake: fetch the links in an inbox and save them as local Markdown.
• Grading and routing: grade each article on two axes — source reliability and information credibility (the Admiralty system). High-value pieces go to project folders with action notes; the rest is parked or discarded; a daily digest is written.
• Human in the loop: fill in a correction table, or just drag a file to another folder; the system turns those corrections into rules it follows next time.
• Knowledge base: compile high-value content into an Obsidian wiki with entities, concepts, topics and two-way links, and lint it regularly for contradictions, orphans and thin pages.

Results

• Seven skills form one pipeline (init, intake, filter, compile, lint, feedback-loop, plus shared config), open source under MIT.
• The screenshot below shows the knowledge base before and after compilation: scattered notes became a connected network.
• The same pattern fits teams: competitor and industry monitoring, support-ticket triage, internal document curation.