Skip to content

Conversation

@piyzard
Copy link
Contributor

@piyzard piyzard commented Aug 31, 2025

closes #129
/claim #129

🎯 Overview

This PR introduces a comprehensive detection rule for AutoGPT Agent Runaway Loop Execution - addressing critical silent failures where agents get stuck in endless planning/research cycles without making meaningful progress. The rule identifies repetitive reasoning patterns that can consume excessive resources and prevent task completion even when left running for hours.

CRE Playground Links

CRE-2025-0169 Playground: Test Rule

📊 AutoGPT Issues Covered

# Issue Type Example Error Pattern
1 Repetitive Research REASONING: I need to research the Volvo V60 turning circle specifications
2 Circular Planning REASONING: I should research and gather information about the exact specifications
3 Loop Detection Warning Loop iteration 8 detected...
4 JSON Parsing Loops Attempting to fix JSON by finding outermost brackets...
5 Endless Goal Analysis REASONING: I need to research and analyze the turning circle in both metric and imperial units
6 Task Never Completion Agent continues research without providing final answer
7 Resource Exhaustion Excessive API calls during endless loops
8 Silent Failure Agent appears active but makes no real progress

🧪 Testing & Validation

image
cat  rules/cre-2025-0169/test.log | preq -r rules/cre-2025-0169/autogpt-runaway-loop.yaml -d

🎬 Demo Environment

Repo link: https://github.com/prequel-dev/cre/tree/main/examples/autogpt-cre-demo

Screencast.from.2025-09-01.14-57-08.mp4

Features:

  • Real AutoGPT simulation with runaway loop patterns
  • Professional AI agent production environment simulation
  • Automated repetitive reasoning log generation with realistic planning cycles
  • Docker containerized environment for consistent testing

References

@piyzard
Copy link
Contributor Author

piyzard commented Oct 22, 2025

cc @Lyndon-prequel

@amanycodes
Copy link
Contributor

amanycodes commented Oct 27, 2025

@piyzard thanks for submitting the rule! if you don't mind could you please share the test repo with me aswell as right now the repo link is pointing to the actual cre repo and it's giving 404. Here's me username: amanycodes

window: 300s
event:
source: cre.log.autogpt
match:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @piyzard mostly LGTM just wanted to mention that the test logs contain all the 6 issues you mentioned but the regex pattern only covers 2 cases:

          - regex: 'REASONING.*I need to.*research'
            count: 3
          - regex: 'REASONING.*I should.*research'
            count: 2

why is that?

- metadata:
kind: prequel
id: AG3xK8mP2qQzYvWr9aLfT7
hash: YpQ9Lm4Zk8TnVb2Ry6HwGs
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hash is not a valid field.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

AutoGPT: Reproduce A High-Severity Failure & Write a CRE Rule [Multiple Winners] [Submit by August 31 11:59 pm ET]

2 participants