One scheduler’s run log held 1,604 records, every one marked a success with a null error field. The same fleet counted from the session registry instead read 1,523 idle, 81 error and 2 interrupted. A monitoring job raised a correct alarm 61 times into a delivery channel that had never been configured, and the run log recorded SUCCESS every time. A watchdog left 47 orphaned processes holding 10.68 GB. None of it was caught by a monitor. Every one passed a liveness check and failed a work check.

For you if: you run more than one agent, on a schedule, without watching each run.

Not for you if: you have not yet run agents in production. Nothing here explains what an agent org, a persona or a delegated session is. It is also not prompt templates, not a Claude Code tutorial, and not code. These are written procedures in prose.

Included: 39 SKILL.md files, 107,549 words, covering running the fleet, judging whether an agent’s report is true, and reporting to the person who owns the org. Copy the folders into your skills directory. They load in Claude Code, Claude.ai, or any agent that reads skill files.

Fourteen days, no questions asked. No reason required, nothing to delete, and nothing to prove. That applies wherever you bought from. Read “Not for you if” above before you buy.

Questions before you buy: support.ashgrovelabs@gmail.com, or support@ashgrovelabs.app which reaches the same inbox. We reply within one business day, Monday to Friday.