What problem does it solve? Running the ai-raccoon MCP memory server in production involves non-obvious operational traps: a 4-hour idle watchdog that silently kills the serve, port-7721 ownership races, CLI verbs that moved under settings, and dotnet tool update no-ops when redeploying same-version builds. This Skill captures measured operational facts and rollout mechanics so deployments and upgrades succeed without rediscovering these failures. ## Core Features & Use Cases - Serve lifecycle management: Start ai-raccoon serve on 127.0.0.1:7721 with --idle-timeout 0 for permanence, background it with logs, and avoid silent 4-hour shutdowns. - Version-rollout watchdog pattern: Cron-driven scripts that gate on version, surgically kill the port owner via lsof -ti :7721, assert promotion settings once with a marker file, and verify release features before marking done. - Local QA deploy protocol: Bump versions uncommitted, rebuild into the local nupkg feed, uninstall/reinstall the global tool to force redeploy, back up the SQLite bank, and run a full manual MCP tool matrix. - Use Case: When upgrading ai-raccoon to a new release, follow the rollout watchdog pattern to update the global tool, restart the serve safely, enable extraction settings once, and verify the maintenance service checkpoints in serve.log. ## Quick Start Ask the agent to check the installed ai-raccoon version, restart the serve on port 7721 with a permanent idle timeout, and verify the maintenance service is checkpointing the WAL.
