What problem does it solve? Managing eval-gated experiments in Caveman Cloud requires strict evidence checks before any lifecycle change, and current agent surfaces are read-only because the control API does not yet enforce lifecycle transitions and evidence gates atomically. This Skill ensures every start, approve, cancel, promote, or rollback request is evaluated against complete evidence and blocked when unsafe. ## Core Features & Use Cases - Evidence-Based Review: Reads experiment state, sample sizes, quality results, guardrails, and cost evidence via MCP or CLI before recommending any action. - Fail-Closed Safety Gates: Refuses approval when results are pending, guardrails are absent, or evidence reports a breach, and never converts experiment lift into verified savings. - Mutation Blocking: Never executes lifecycle mutations, even after user approval, because agent-generatable action strings are not proof of human intent. - Use Case: A user asks to approve experiment 7f after seeing positive lift. The Skill reads the experiment detail and results, finds a missing latency guardrail, reports "evidence incomplete", and blocks the approval instead of executing it. ## Quick Start Ask the assistant to review Caveman experiment 7f and recommend whether it should be approved, started, cancelled, or rolled back.
