What problem does it solve? Long coding sessions drift, and a single executor model can lock onto a weak approach without any checkpoint. This Skill binds one high-tier advisor (the built-in advisor tool or a warm pstack:poteto-agent) and consults it at key moments so plans, corrections, and stop signals arrive before costly mistakes. ## Core Features & Use Cases - Session-scoped advisor binding: Detects the host (Claude, Codex, or third-party), walks the model floor, and binds the strongest available advisor tier once per session, with fail-closed behavior when the bind fails. - Structured consult cadence: Consults after orientation and after writes and validation, following the hard rule of calling the advisor before the first write, edit, or state-changing shell call. - Evidence recording: Writes consult evidence (selected tier, fallback reason, reply path, validation, unresolved risks) to a session-controlled model-tier-run.json record. - Use Case: Before refactoring a module, the session consults the bound advisor with a brief built from the consult-format reference, receives an ENDORSE / CORRECTION / PLAN / STOP signal, and proceeds or adjusts accordingly. ## Quick Start Ask the agent to bind a team advisor for this session and consult it before making the first code change.
