Survey Generator

You point it at a public anchor resource, it extracts a taxonomy and real paper citations, then sends everything to your choice of LLM (Anthropic, OpenAI, OpenRouter, Fireworks, custom) to write the prose. Output lands in your wiki's `derived/surveys/` with full bibliography rows appended to `sources.md`, so the research loop can build on it later. Good for onboarding a new domain or synthesizing what you've already collected. The workflow enforces real citations only and lets you compare providers by running the same bundle through different models. It's provider-agnostic by design, which matters when you want reproducible research artifacts that aren't locked to one vendor.