Dossier Collect

You point it at something,a username, file, ADR, symbol, or concept,and it fans out recursively across web search, memory, embeddings, git, and codebase to build a provenance-tracked graph of connected entities. It's BFS expansion with budget controls, so you get a JSON graph plus a Markdown summary with Mermaid diagrams. Use it when you need to map everything related to a seed rather than answer a specific question. The parallel fan-out approach is smart: it fires all source queries in one message per round instead of waiting sequentially. It also de-dupes by embedding similarity to avoid redundant nodes. Good for onboarding, impact analysis, or tracking down who owns what across a sprawling project.