What problem does it solve? Architecture decisions made in conversation get lost, re-debated, or built on stale assumptions. This Skill enforces a structured, evidence-based session that verifies current project state, audits multiple references before generalizing, and persists every decision as an ADR before the session ends. ## Core Features & Use Cases - Six-phase workflow (A-F): state verification, multi-reference audit (Regla del 3 gate), constraint formalization, options with rejection list, in-session ADR persistence, and recap via handoff/context-refresh skills. - Blocking quality gates: refuses to advance past the audit phase without at least two references for cross-instance decisions, and refuses to end without writing the ADR file. - Socratic option filter: evaluates 3-5 candidate options against a formalized constraint table using invariant checks, naive-trap analysis, falsifiable rejection criteria, and one-way vs two-way door classification. - Use Case: A team must choose a shared SDK template across multiple sensor products. The session audits two-plus reference implementations, formalizes constraints, records the decision as an ADR in the repo's docs/adr directory, and updates the plan and GitHub issues in the same session. ## Quick Start Ask the agent to run an architecture session for your project, for example by saying "run an architecture session to decide the config schema for the sensor SDK platform".
