run

Run a Claude Code-style dynamic workflow. Plan the task, fan it out across parallel subagents (workflows:worker), independently re-verify every finding (workflows:verifier), attack the result adversarially (workflows:adversary), iterate to convergence, and deliver one coordinated report. Invoke with /workflows:run <task>. Use for large, parallelizable, verifiable work: codebase audits, migrations, broad research, plan stress-testing.