Results clarify when causal cross-query memory improves repair and when broader memory representations remain preferable, and show that negative memory contributes modestly, the value of type conditioning and lexical-dense ranking is dataset dependent, and schema-local experience provides the most consistent benefit.
Abstract
LLM agents that repair failures often discard successful corrections, forcing later episodes to rediscover similar solutions. We study whether finalized repair outcomes can improve subsequent Text-to-SQL episodes without parameter updates. We introduce MERIT, a training-free agent that maintains an online dual-polarity memory of oracle-verified corrections and observed unsuccessful directions. Under oracle-assisted benchmark feedback, only memories from earlier finalized episodes are eligible for retrieval. A deterministic classifier assigns a coarse failure type, which conditions a hybrid lexical-dense retriever before the frozen model generates each revision. Using Qwen2.5-7B-Instruct with identical initial predictions and repair budgets, \method{} improves execution accuracy over stateless iterative repair from \(66.34\%\) to \(69.79\%\) on Spider and from \(47.35\%\) to \(48.44\%\) on BIRD. Paired analyses provide clear evidence for the Spider gain but weaker evidence on BIRD. MERIT is not reliably separated from untyped dynamic retrieval on either benchmark, while Reflexion-style memory reaches \(51.24\%\) on BIRD at substantially higher inference cost. Ablations show that negative memory contributes modestly, the value of type conditioning and lexical-dense ranking is dataset dependent, and schema-local experience provides the most consistent benefit. These results clarify when causal cross-query memory improves repair and when broader memory representations remain preferable. Our implementation is available here:
PMCoder is presented, an issue-resolution agent that couples a hierarchical phase planner with episodic memory that outperforms either component alone and reduces repeated failed actions, empty-patch exits, and context-window exhaustion.
It is shown that feedback is a sufficient signal for continual learning when the frozen model is paired with an external memory that distils each episode into retrievable natural-language rules when the frozen model is paired with an external memory.
V. Tablan, Scott Taylor, Kristoffer Bernhem· arXiv.org· 0 citations
It is shown that evidence-gathering should be evaluated as a trajectory-level control problem, separately from answer-side reasoning, because larger hidden thinking budgets do not necessarily increase evidence inspection.
Experiments establish ReTree as an effective self-correcting memory abstraction for long-horizon search, and show that ReTree consistently outperforms Full-Trajectory ReAct in question-answering and search benchmarks.
Aijun Yang, Qianxue Guo, Ziyi Huang et al.· 0 citations
As language agents persist and collaborate over long horizons, a stored fact is no longer disposable context: once recalled, it steers tool calls, planning, and cross-agent agreement. We argue that memory reliability breaks down into two distinct failure modes: staleness (the world changed after a fact was stored) and false consensus (agents copied a wrong value until it looked agreed). Better retrieval fixes neither. VAIL is a governed-memory architecture with one rule: retrieval authorizes inspection, not use. Two independent gates enforce it. An executable time-to-live gate re-runs a stored check against the live environment instead of trusting an age threshold; a private append-only ledger keeps conflicting observations visible and settles trust from outcomes rather than from how often a value was repeated. On a controlled multi-agent benchmark with thirty paired seeds and real model calls, each gate clears its own failure metric while leaving the other untouched, and only the combination collapses task failure, by about 84 percent, with every comparison surviving Bonferroni correction. VAIL outperforms faithful A-MEM and Mem0 ports under a uniform memory-injection budget, stays neutral on trap-free public benchmarks, degrades gracefully as trap pressure rises, and replicates under a second controller model. Two studies built from real GitHub release histories validate each gate in the wild: re-running stored checks repairs real environment migrations, and outcome settlement overturns wrong majorities produced by real version skew.
Ming Wang, Ke-Yang Han, Ru-Yi Wang et al.· 2026 12th International Conf...· 0 citations
Language agents increasingly rely on reusable skills, but post-failure repair is often handled by opaque one-shot reflection: a model generates a skill patch without explicitly maintaining how failure explanations relate to candidate repairs or how unsuccessful retests should influence later edits. We introduce RESKILL, a structured repair framework that maintains an explicit repair state across repair rounds. Given a failed rollout, the framework links failure hypotheses to candidate skill patches, selects local repairs through coverage-based attribution, retests the edited skill set in the environment, and uses retest outcomes to guide subsequent repair updates. The language model supplies structured repair factors, while the repair procedure records them, compares local skill patches by how well they address active failure explanations, and carries unsuccessful retest outcomes into later repair rounds. We evaluate RESKILL on ALFWorld and TextCraft across three model sizes under fixed repair budgets. RESKILL obtains the strongest final success in all six benchmark-model settings, improving average final success by 3.7 percentage points over direct repair and 3.3 points over hypothesis-conditioned repair. These results suggest that explicit attribution alone is insufficient; durable improvement emerges when attribution is integrated with repair selection and persistent retest-conditioned update.
Meng-Yi Deng, Xin Li, Du-Yi Pan et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
