AI Agent Frameworks / Scenario
The Chat That Would Not Die
An AutoGen pair burns 50 turns because nobody's termination check matches what the model actually says. Cheap bug, expensive invoice.
- Difficulty
- Medium
- Format
- Scenario
- Points
- 150
- Estimate
- 12 min
// MISSION BRIEF
Your Mission
A nightly AutoGen job (assistant + user proxy running the test suite via a registered tool) sometimes completes in 6 turns, sometimes burns all 50 and gets force-stopped. Finance noticed before engineering did.
Read the configuration and the transcript, find why termination fails, and design the tests that keep conversations mortal.
// FIRST CONTACT
Battle teaser
First artifact
nightly_fixer.py
Why did the conversation never terminate on its own?
- AAutoGen ignores is_termination_msg when tools are registered
- BThe assistant never tried to terminate
- Cis_termination_msg uses exact equality against "TERMINATE", but the model emits surrounding text ("... TERMINATE." with prose and punctuation), so the predicate never matches and the loop runs to the cap
- Dmax_consecutive_auto_reply=50 disables termination checks
Answers, scoring, hints, and the full battle stay sealed.
// SKILL TAGS
autogenterminationtool-usemulti-agentai-agents