independent-consult

Runs a blinded, independent consult on the patient's records, the kind of fresh outside read a family might hire. A team of agents that has not seen the treating team's conclusions or the family's prior analysis builds a fact base, generates every hypothesis the records cannot yet rule out, investigates each family of hypotheses with literature, attacks its own conclusions with a red team, reconciles against the treating team and prior analysis, and produces one verified, cited report with a test plan. Use only when someone explicitly asks for an "independent consult", "blinded second opinion", "fresh unbiased review", or "leave no stone unturned", or for a consult addendum or status. Expensive: from about 10 to 35 or more agents.

License

MIT

Compatibility

Best in Claude Code or Cowork, which support a blind subagent type through this plugin's blind-reader agent. Other hosts can run the blind stage in a separate session opened in the snapshot folder. Python 3.10+.