What problem does it solve? Manually building a knowledge base requires repeated searching, reading, deduplicating, and writing. This Skill automates that cycle: it pops research seeds from a queue, fetches sources, drafts wiki pages, dedupes claims against existing pages, and enqueues follow-up topics until a budget, depth, or convergence limit halts the run. ## Core Features & Use Cases - Autonomous BFS research loop: Pops seeds breadth-first from a SQLite-backed queue, fetches sources via pluggable fetchers (web via DuckDuckGo Lite, arXiv API, GitHub search), compiles pages, and enqueues follow-up seeds up to a max depth. - Cost and safety controls: Halts on USD budget cap, max-pages cap, depth cap, convergence detection (Jaccard novelty below 5% for 3 consecutive pages), a kill-switch file, or privacy guard that blocks non-local fetchers for private wikis. - Auditable output: Each run writes a human-readable Markdown log and a structured JSON stats file under the wiki root, and every page is upserted through the wiki store so FTS5 search stays consistent. - Use Case: Tell the agent to "research memory consolidation in agents" — it seeds the queue, runs the loop overnight, and you wake up to a grown wiki with sourced claims and follow-up pages. ## Quick Start Ask the agent to research a topic by saying "research <topic>" or "grow the <slug> wiki", and it will seed the queue and run the budget-capped loop automatically.
