monitor-rollout

Runs a change-monitoring plan inline in the current coding-agent session. Foreground, single-issue, multi-environment.

When to use

Only after a monitoring plan exists and the change has been merged or is about to deploy. Specifically:

• The user just merged a PR and wants production watched.
• The user is doing post-deploy validation, watching a canary, or babysitting a rollout.
• The user pasted a path to a monitoring plan file.
• The user invoked this directly from rollout after the planning phase.

This skill runs in the main session, not as a background subagent. The user wants to see the polling and the checkpoints happen — visibility is the UX. Between checkpoints, the skill kicks off a background sleep_until.sh that emits a [RESUME …] notification at the next checkpoint time; the agent reads the notification and resumes the loop.

Operation contract

Input: path to a monitoring plan file produced by plan-rollout. The plan section's structure is documented in plan-rollout/assets/monitoring-plan-template.md.

Lifecycle: stays in the main session for the entire monitoring window. Each checkpoint is a sequence of tool calls; the user reads the resulting status report inline as a normal assistant message. After emitting a checkpoint report (and verifying no terminal condition was reached), the skill invokes bash monitor-rollout/scripts/sleep_until.sh <abs-time> <plan-path> <offset> via the Bash tool with run_in_background: true. When the sleep completes the script echoes a [RESUME …] line, which arrives in the agent's chat as a notification — the agent picks it up and resumes the next checkpoint.