loop-runtime-monitor

Detect stalled agent loops, repair churn, and budget overruns from run evidence.

1|Updated Jun 20, 2026

Search Tags:#stall-detection #agent-runtime #state-json #loop-monitoring #repair-churn #budget-overrun #runlog

Ready to run

npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-runtime-monitor

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: loop-runtime-monitor
Source: https://github.com/SollanSystems/loop-engineer/tree/main/skills/loop-runtime-monitor
Command: npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-runtime-monitor

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill watches a running agent loop from the outside and tells you when it has stopped making real progress, started churning repairs, or is about to run out of budget.

Core Features & Use Cases

Stall Detection: Spots repeated work on the same task with no measurable score improvement.
Repair Churn Detection: Flags repair attempts that keep failing without improving the loop's best score.
Budget Watch: Detects when a numeric budget axis reaches zero or below and escalates to a human approval step.
Operational Recommendation: Returns one clear next move such as continue, replan, revert, approval, or terminate.
Use Case: A long-running coding agent appears busy but keeps repeating the same task; this Skill helps you intervene before the run burns the remaining budget.

Quick Start

Ask this Skill to monitor a loop directory and report whether the run is still progressing, what signal fired, and what intervention you should take next.

Frequently Asked Questions about loop-runtime-monitor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I detect if my coding agent loop is stuck or spinning without making progress?▼

Stall detection identifies stuck agent loops by checking external run evidence for repeated tasks with no score improvement. It parses state files to spot loops that are spinning without making real progress.

What is repair churn in agent runtime monitoring and how do I flag it?▼

Repair churn detection flags repeated repair attempts that keep failing without improving the loop's best score. It monitors state and ledger files to alert you when fixes are unproductive.

How do I monitor an agent loop for budget overruns?▼

Budget watch detects when a numeric budget axis reaches zero or below and escalates the run to a human approval step. It monitors state files to prevent an agent loop from burning remaining resources.

Can I monitor a long-running agent loop that exposes state.json and runlog files?▼

Yes, loop monitoring applies to long-running runs that expose .loop/state.json and RUNLOG.md for live monitoring. It requires deterministic parsing of these files to evaluate whether the loop is progressing or nearing exhaustion.

How do I get an operational recommendation for a stalled agent loop?▼

The monitor returns one advisory recommendation mapping to continue, replan, revert, approval, or terminate. It applies conservative threshold checks to state and ledger files to determine the best next intervention.

What are the limitations of using automated stall detection for agent loops?▼

Stall detection requires runs to expose .loop/state.json and RUNLOG.md files for deterministic parsing. It cannot monitor agent loops lacking these external run evidence files.