What problem does it solve? Deciding whether to pursue an idea often gets reduced to a single unchallenged opinion or polite consensus. This Skill forces genuine disagreement between independent subagents in opposing roles, grounded in real repository facts, and produces a synthesized verdict instead of a list of opinions. ## Core Features & Use Cases - Grounded Debate Setup: Investigates actual scale, consumers, and blast radius in the repo before any position is assigned, so arguments rest on facts rather than assumptions. - Opposing Role Assignment: Spawns Advocate, Skeptic, and optional Migration/Risk subagents in parallel with an identical brief, ensuring differences come from role rather than unequal information. - Schema-Validated Verdicts: Synthesizes a proceed, do_not_proceed_for_now, or proceed_with_modification verdict, validates it against a JSON schema, and persists it as a finding or known-issue with a concrete revisit trigger. - Use Case: Before committing to replacing hand-rolled retry logic with a library, run a debate that investigates call sites, assigns opposing reviewers, and delivers a decided recommendation with the losing arguments recorded. ## Quick Start Ask the agent to run a design debate on whether the team should adopt a specific change, giving the precise A-vs-B decision you want stress-tested.
