SAFARI: Scaling Long Horizon Agentic Fault Attribution via Active Investigation
For developers debugging complex multi-step agent failures, SAFARI provides a scalable solution that overcomes context window limitations.
SAFARI introduces a tool-augmented diagnostic loop with short-term memory to scale fault attribution in long-horizon agentic tasks, outperforming state-of-the-art by 20% on Who&When and 19% on TRAIL GAIA while maintaining 0.58 precision even 5x beyond context limits.
As autonomous agents tackle increasingly complex multi-step, multi-agent tasks, their execution trajectories have scaled beyond the constraints of even the largest context windows. Current methods for effectively diagnosing agent failures load the full trajectory into an LLM's context window, which suffers from attention dilution and fails when agentic traces inevitably exceed context limits. To address this, we introduce SAFARI (Scaling long-horizon Agentic Fault AttRibution via active Investigation), a framework that replaces linear context loading with a tool-augmented diagnostic loop. By equipping LLMs with a specialized toolbox to read and search trajectory segments alongside a persistent Short-Term Memory (STM) for cross-turn reasoning, SAFARI effectively decouples diagnostic accuracy from architectural context limits. Our experiments demonstrate that SAFARI outperforms state-of-the-art results by 20% on the Who&When dataset within a 1M token budget, and by 19% on TRAIL GAIA subset on a 25K token budget. Most significantly, SAFARI maintains a 0.58 precision even when the target fault resides 5x beyond the model's native context window, a scenario where traditional evaluators fail entirely.