Structured decision-making — surface the goal, make an initial call, stress-test with two adversarial subagents, then render a final verdict
Make a clear, reasoned decision from context — no user input during the process. Works for personal and technical decisions.
If the conversation is thin (topic stated in a single message with no elaboration, constraints, or history), spawn a context-gathering subagent. It reads relevant files — vault notes, project docs, recent daily notes, code — and returns: goals at stake, constraints, and what's been tried or ruled out. Synthesize before proceeding.
State what is being chosen between, what outcome is being optimized for, and what constraints apply. If ambiguous, infer the most plausible interpretation — do not ask.
Commit to a clear, specific decision:
Spawn two subagents in parallel:
Steelman: Build the strongest case FOR the initial decision — evidence, analogies, logic. Ignore counterarguments.
Red Team: Build the strongest case AGAINST it. Assume it's wrong; find the best alternative and surface risks, overlooked factors, failure modes.
Each returns: a 2–3 sentence argument and, if overturning, an alternative decision with reasoning.