Market Recon fans out a coordinated agent team — dimension analysts, a scout, a buyer-side researcher, and a fact-checker — then builds a world model that gets smarter with every run.
Plain language, a capability area, or a pasted PRD. Market Recon normalizes the input, checks scope, and asks two questions: your role and the decision you're trying to make.
Dimension analysts research each capability lane in parallel. A scout finds what they miss. A buyer analyst reads G2, Reddit, and forums. A fact-checker resolves citations before anything gets written.
A world model records every player and angle. The next run inherits it — the scout always searches past what's known. After Run 4, a longitudinal brief synthesizes market movement across all runs.
claude --version
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 to your ~/.claude/settings.json
To make /market-recon available in every project, run ./install.sh from the repo root.