What problem does it solve? Changes in the AI-DEMO2 repo often appear correct, tested, and merged yet never actually run in the live Docker stack, and worktree-based test runs fail for environmental reasons that look like code bugs. This Skill documents the verified operational gotchas so you can prove a code path executes, deploy changes correctly, and avoid misdiagnosing infrastructure noise as regressions. ## Core Features & Use Cases - Worktree test setup: Bootstrap node_modules in fresh git worktrees via scripts/bootstrap-worktree.sh and avoid the jest --testPathIgnorePatterns trap that drags live-stack suites into unit runs. - Docker deploy correctness: Distinguish bind-mounted services (restart suffices) from image-built services (rebuild required), including the dual scope-topology.json copies needing different deploy verbs. - Live verification discipline: Prove code paths run via container log markers, verify deploys by content rather than stamps, and detect when a concurrent agent session's deploy invalidated your live run. - Use Case: After merging a topology change, you grep your own diff marker inside the authz-server container, rebuild only the image-based services, and restart ping-gateway — instead of shipping half the change while every container reports healthy. ## Quick Start Ask the agent to verify whether your merged change is actually running in the AI-DEMO2 Docker stack and follow the worktree and deploy guidance it applies.