What problem does it solve? Running Foundry agent optimization jobs involves billable submissions, baseline configuration drift, and the risk of breaking native MCP tool execution. This Skill records the constraints and command semantics to check before invoking Castia optimizer commands, so candidates are applied and deployed correctly. ## Core Features & Use Cases - Optimizer command semantics: Distinguishes free offline operations (optimize --check, optimize run --dry-run) from billable job submission, status polling, cancellation, and candidate application. - Baseline configuration guidance: Defines the canonical .agent_configs/baseline/ metadata shape, pure tool provider requirements, and load_agent_config anchoring rules. - Native MCP preservation: Shows how to keep upstream tool names like web and search_query, pass tokens at request time via Model.respond_with_tools, and avoid replacing remote schemas with local wrappers. - Use Case: Before submitting a billable optimize run job against a deployed Responses-only agent, use this Skill to verify the dry-run preview, confirm the wire contract, and plan the azd deployment handoff after applying a candidate. ## Quick Start Ask the assistant to review the constraints for running a Castia optimizer dry-run against my Foundry agent's baseline configuration before submitting a billable job.
