loop

name	loop
description	Schedule recurring follow-up messages into the current conversation. Supports fixed-interval monitoring and one-shot wakeups for dynamic self-pacing.
metadata	{"author":"CoC","version":"0.0.2"}

Loop — Recurring Follow-Ups

Schedule recurring follow-up messages into the current conversation so the AI can monitor, check, or re-evaluate on a cadence without human intervention.

When to Use

• The user asks to "check back", "monitor", "keep an eye on", or "remind me".
• A task needs periodic re-evaluation (build status, deployment health, metric tracking).
• The user wants dynamic self-pacing: "come back when X is ready".

Available Tools

When this skill is active you have one additional tool: loop, with an action parameter:

Action	Purpose
loop action create	Create a fixed-interval recurring loop (requires description, interval, prompt). First tick fires after one full interval — the current turn is the implicit first run.
loop action cancel	Cancel an active or paused loop by loopId.
loop action list	List all loops for this conversation, optionally filtered by status.

The scheduleWakeup tool (one-shot delayed follow-up) is always available regardless of this skill.

Interval Parsing

Intervals accept human-friendly strings: 30s, 5m, 1h, 2h, 1d, or raw milliseconds. Minimum interval for loop action create is 10 seconds. Minimum delay for scheduleWakeup is 1 second.

Choosing Between Loop and Wakeup

Scenario	Tool
Periodic monitoring (every 5 min check build status)	loop action create
One-time delayed check ("check in 30 minutes")	scheduleWakeup
Dynamic pacing ("come back when the deploy finishes")	scheduleWakeup — check once, then schedule another if not done

Slash-Compatible Fixed Interval Mode

When this skill was explicitly selected and the user message starts with a duration followed by a task, treat it as a request for a recurring fixed-interval loop.

Examples:

• 1m what's the time now? -> create a loop every 1 minute with prompt what's the time now?
• 30s check the build -> create a loop every 30 seconds with prompt check the build
• 2h remind me to stretch -> create a loop every 2 hours with prompt remind me to stretch

In this mode:

1. Run or answer the prompt immediately in the current turn.
2. Call the loop tool with action create, the parsed interval, and the remaining prompt.
3. Do not call scheduleWakeup; that tool is for one-shot delayed follow-ups.
4. Use the default TTL unless the user specifies a duration or stop condition.
5. If the remaining prompt is empty or nonsensical, ask for clarification instead of creating a loop.

User Confirmation

Before creating a loop, always confirm with the user:

1. What you will monitor and why.
2. The proposed interval and how long it will run.
3. The stop condition — when you will cancel the loop.

For explicit fixed-interval slash-compatible input such as 1m check status, the user's command is the confirmation. Do not ask an extra confirmation unless the interval, task, or stop condition is ambiguous or risky.

Example confirmation:

I'll check the CI pipeline status every 5 minutes for up to 3 hours, and stop once all checks pass or a clear failure is detected. Shall I set this up?

Intent-Based Escalation

For very long intervals (e.g. hours or days), consider whether the schedule system might be more appropriate. Mention this to the user as a suggestion — do not enforce it.

Stop-Condition Recognition

Watch for signals that a loop should end:

• The monitored condition is met (build passed, deploy complete).
• The user says "stop", "cancel", "enough", or "no more".
• Repeated identical results suggest nothing is changing.

When a stop condition is detected, cancel the loop with the loop tool (action cancel) and summarize the outcome.

Circuit Breakers

The system enforces safety limits automatically:

• 3 consecutive failures → loop auto-pauses.
• 3-day TTL default → loop expires (override with the ttl parameter).
• Max 100 consecutive automated turns per process → pauses to prevent runaway loops.
• Max 50 active loops per server.

You do not need to enforce these — they are handled by the runtime.

Best Practices

• Keep loop prompts focused and specific. A good loop prompt is a clear instruction, not a vague "check things".
• Set an appropriate TTL. Don't leave loops running indefinitely.
• Prefer shorter intervals for active debugging (30s–2m) and longer intervals for background monitoring (5m–1h).
• Use the loop tool with action list to show the user their active loops when asked.
• When a loop detects the goal is met, cancel it immediately rather than waiting for the next tick.