What problem does it solve? Deciding what to improve is often harder than building it. This Skill produces a ranked, evidence-grounded ideation artifact by generating many candidate ideas, critiquing them adversarially, and presenting only justified survivors instead of a naive brainstorm list. ## Core Features & Use Cases - Grounded Multi-Agent Ideation: Dispatches parallel grounding agents (codebase scan, learnings search, web research, issue intelligence) and 4-6 ideation sub-agents across six divergent frames to produce 25-30 deduplicated candidates. - Adversarial Filtering with Warrants: Every surviving idea must carry an articulated warrant (direct evidence, external prior art, or reasoned argument) and pass an ambition floor; rejected ideas each get a one-line reason. - Mode-Aware Routing: Classifies the subject as repo-grounded, elsewhere-software, or elsewhere-non-software, adapting grounding, generation frames, and persistence (local docs/ideation/ file vs Proof review) accordingly. - Use Case: Ask "what should I improve in the authentication system" and receive a ranked ideation document with 5-7 surviving ideas, each with warrant, rationale, downsides, confidence, and complexity, plus a rejection summary. ## Quick Start Ask the agent to ideate on a focus area, for example: give me ideas to improve the onboarding flow.
