AgentCollabBench: Diagnostic Benchmark for Multi-Agent LLM Systems

AgentCollabBench is a diagnostic benchmark dataset for multi-agent LLM systems, created by AgentCollabBench and last updated on 2026-05-06. It targets process-level failures that single-agent benchmarks cannot expose, such as failures emerging from inter-agent communication.

Use Cases

• Diagnosing constraint decay under peer pressure based on the described failure mode.
• Measuring multi-hop information loss in agent communication chains.
• Analyzing false-belief propagation between agents.
• Testing for private context leakage in collaborative scenarios.

Strengths

• Specifically designed to expose failures unique to multi-agent collaboration.
• Targets four described process-level failure modes: constraint decay, information loss, false-belief propagation, and context leakage.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AgentCollabBench

Collection Method

Likely contains constructed scenarios or tasks to diagnose specific multi-agent failure modes.

Time Range

null

Freshness

Last updated 2026-05-06 07:10:06; freshness should be verified.

Geography

null