Sentry improves large language agent recovery from task failures
Sentry: Learning to Recover from LLM Agent Failures at Test Time
Machine LearningArtificial IntelligenceComputation and Language
Summary
Large language model (LLM) agents sometimes fail during tasks because of errors like calling tools incorrectly or getting stuck repeating actions. The authors found that how failure knowledge is shared with the agent is as important as what it contains. Their system, Sentry, stores failure lessons separately and only uses them when needed to help the agent recover. This approach improves agent performance significantly on several tests, and lessons learned can help with new tasks too.
What this means in practice
- •For ai system developers: Enhance reliability of AI agents by integrating a failure management layer that aids in recovery without bloating the agent’s operational context.
- •For customer support platform teams: Improve automated agent responses by enabling dynamic failure handling informed by past recovery lessons stored externally.
Authors
Changxiu Ji, Amy Lu, Qizheng Zhang, Kunle Olukotun
Abstract
LLM agents often fail mid-task due to invalid tool calls, repeated actions, or poorly grounded reasoning, and learning from these failures is a path to reliability. We find that how failure knowledge reaches the agent matters as much as what it contains. Failure lessons are conditional: kept in the agent's context, they misfire when their failure is absent, and removing them from an evolving playbook improves performance. Runtime interventions, in contrast, act only when a failure occurs but do not learn from their repairs. We argue that failure knowledge is conditional knowledge and should be conditionally exposed, and instantiate this principle in Sentry, a failure-management layer that runs alongside the agent. When Sentry detects a failure, it retrieves matching lessons from an external playbook to guide recovery, verifies without access to task rewards whether the agent recovered, and stores a new lesson only if it did; the full playbook never enters the agent's context. Across multiple agentic benchmarks, Sentry outperforms the strongest runtime-intervention baseline on every benchmark, by 37\% on average, and the strongest context-evolution baseline by 39\% on the two benchmarks where both are evaluated; combining Sentry with context evolution yields further gains. Learned lessons transfer to held-out tasks, and controlled experiments show that exposing the full playbook to the agent lowers performance even when relevant lessons remain available on demand.