monitor

What problem does it solve? Long-running jobs like CI runs, deployments, remote training jobs, and multi-module builds require constant babysitting. This Skill starts the built-in Pi monitor workflow to observe the target, take safe authorized actions, and finish the goal without manual polling. ## Core Features & Use Cases - Continuous Observation: Each cycle runs a read-only observe step that checks goal completion, active work, failures, and blockers before deciding to wait, act, or stop. - Authorized Action and Recovery: Performs direct advance and recover actions, and composes plan-change plus Autoimplement workflows for code or configuration repairs, all strictly within existing authority. - Cost and Authority Guardrails: Verifies spending approvals, cost ceilings, and protected contracts before any paid work, and stops when limits are unclear or exceeded. - Use Case: Ask the agent to watch a remote training job that must produce 27 verified module outputs; the monitor resumes failed workers, refreshes launch files, reports progress every 15 minutes, and stops when done or blocked. ## Quick Start Ask the agent to monitor your running deployment or job and finish the goal, for example: monitor the CI run on my repository and resume any failed jobs until all checks pass.