From General Agents to RCA Experts: A Self-Evolving Harness for Root Cause Analysis
Ranking
Observed public metrics from 1 member.
Merged summary
TL;DR - OpsHarness is a self-evolving harness that adapts general-purpose LLM agents for root cause analysis by converting prior diagnostic experience into reusable, system-specific expertise. It reports 59.0% top-1 accuracy across public benchmarks and an industrial deployment, outperforming both an unmodified general agent and specialized RCA baselines.
- The study finds modern general agents often outperform purpose-built RCA agents, but their external adaptation harness remains a key production bottleneck.
- OpsHarness combines layered operational knowledge and an “idea-card” tool library with workflows for setup, diagnosis, evolution, and verification.
- It learns by contrasting successful and failed diagnostic trajectories, extracting atomic update proposals, and applying dual-gate verification to limit overfitting and regressions.
- Its 59.0% top-1 accuracy represents a 63.4% improvement over a bare general agent and a 4.02× improvement over baseline RCA agents.
Sources (1)
From General Agents to RCA Experts: A Self-Evolving Harness for Root Cause Analysis
TL;DR - OpsHarness is a self-evolving harness that adapts general-purpose LLM agents for root cause analysis by converting prior diagnostic experience into reusable, system-specific expertise. It reports 59.0% top-1 accuracy across public benchmarks and an industrial deployment, outperforming both an unmodified general agent and specialized RCA baselines.
- The study finds modern general agents often outperform purpose-built RCA agents, but their external adaptation harness remains a key production bottleneck.
- OpsHarness combines layered operational knowledge and an “idea-card” tool library with workflows for setup, diagnosis, evolution, and verification.
- It learns by contrasting successful and failed diagnostic trajectories, extracting atomic update proposals, and applying dual-gate verification to limit overfitting and regressions.
- Its 59.0% top-1 accuracy represents a 63.4% improvement over a bare general agent and a 4.02× improvement over baseline RCA agents.