What problem does it solve? Choosing which AI model runs each pstack role (bug-fix, arena runners, swarm workers, and more) is tedious and error-prone when done by hand, and guessing model slugs produces broken configurations. This Skill detects the models actually available in your session and writes a validated, always-applied Cursor rule that overrides pstack's defaults. ## Core Features & Use Cases - Model Detection and Validation: Enumerates model slugs available to Task subagents and refuses to write any slug it cannot confirm, falling back to a user-pasted list when detection fails. - Budget-Based Effort Mapping: Applies one of four budgets (unlimited, large, medium, small) by adjusting the reasoning effort token of every role's model, with family-aware fallback when a target slug is unavailable. - Idempotent Rule Writing: Overwrites ~/.cursor/rules/pstack-models.mdc with alwaysApply: true, so re-runs update the configuration cleanly instead of appending duplicates. - Use Case: You want pstack's arena runners on a cheaper reasoning tier. Run the setup, pick the small budget, confirm the role table, and the rule is written so every future session uses the new mapping. ## Quick Start Ask the agent to run /setup-pstack to detect your available models and configure pstack's per-role model choices and reasoning budget.
