What problem does it solve? Verifying that sdd-lite orchestration changes (prompts, wrappers, routing, handoffs) actually behave correctly in real CLI sessions is hard to do without contaminating the evaluation context. This Skill runs behavioral probes in fresh child sessions of the active CLI so regressions are caught with clean, unbiased evidence. ## Core Features & Use Cases - Isolated child-session probes: Evaluates triggers, wrappers, routing, worker bypass, resume, approvals, and main stage prompts in fresh sessions of the active provider CLI. - Cost preview before execution: Runs a dry-run preview showing probe names and child-session counts, requiring explicit confirmation before executing paid runs. - Structured analysis reports: Produces a normalized analysis YAML covering provider-specific prompt incompatibilities, route deviations, isolation failures, and recommendations, then attaches it to regenerate reports and history. - Use Case: After modifying an orchestrator prompt or routing rule, run a smoke-profile Level 2 evaluation against your active provider to confirm triggers fire correctly and worker bypass isolation still holds before shipping the change. ## Quick Start Run a Level 2 smoke evaluation for my workspace and case against the active provider, show me the cost preview first, then execute and attach the completed analysis.
