Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.28834

Where Should a Stuck Agent Spend Its Next Tokens? Causal Attribution of Inference-Time Resources

Saiyue Lyu·Jerrold Huang·Yunbei Zhang·Janet Wang·Zhenting Wang·Yingqiang Ge·Haibo Zhang
View original source ↗Read source document ↗Export metadata ↓

Abstract

Inference-time scaling is often treated as a token-budget problem, but a failed coding agent can seek qualitatively different resources: computation, information, or feedback. We introduce Banzhaf Resource Attribution (BRA), a controlled factorial framework that measures each resource's marginal contribution to repair and its interactions while holding the failed state and revision protocol fixed. We evaluate six models on 194 function-level failures from MBPP+ and HumanEval+ and 126 repository-level failures from SWE-bench Verified. Task-relevant specifications consistently improve repair, whereas public-test feedback provides small gains and retrieved code little to none. The value of additional computation declines with model capability, while information remains useful; resource channels largely exhibit diminishing returns, although complementary information can reinforce itself. Yet three of four frontier models select feedback in 95-100% of cases and obtain lower gains than uniform random resource selection. BRA thus exposes a gap between the resources agents prefer and those that measurably improve repair.

Explore related subjects

cs.SEcs.AIcs.CL

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Saiyue Lyu, Jerrold Huang, Yunbei Zhang, Janet Wang, Zhenting Wang, Yingqiang Ge, Haibo Zhang. 2026-10-20. Where Should a Stuck Agent Spend Its Next Tokens? Causal Attribution of Inference-Time Resources. https://arxiv.org/abs/2610.28834

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Generating Mitigations for Downstream Projects to Neutralize Upstream Library Vulnerability

Third-party libraries are essential in software development as they prevent the need for developers to recreate existing functionalities. However, vulnerabilities within these libraries pose significant risks to dependent projects. Upgrading dependencies is infeasible when no patched version is available or when downstream projects have strict version constraints. Moreover, repairing the vulnerability proves challenging when the source code of the library is inaccessible. State-of-the-art automatic vulnerability repair and automatic program repair methods both fail to address this issue. Therefore, mitigating library vulnerabilities without access to source code or available patches is crucial for a swift response to potential security attacks. Existing tools encounter challenges concerning generalizability and functional security. Upon vulnerability disclosure, LUMEN prepares a pool of candidate mitigation strategies by retrieving explicit workarounds from public sources and supplementing the pool with generated strategies based on vulnerability reproduction behavior. LUMEN then selects a suitable strategy, autonomously gathers repository-specific context, and adaptively generates a validated mitigation patch. We evaluate LUMEN on 474 impacted functions associated with 94 vulnerabilities. LUMEN successfully mitigates 393 functions, achieving a success rate of 82.91%, which substantially outperforms the adapted baseline in blocking vulnerability reproduction under our exploit-based evaluation. Additionally, our ablation study evaluates the contributions of strategy composition, strategy selection, and repository-level context collection. Further evaluations show that LUMEN can mitigate vulnerabilities disclosed after the knowledge cutoff of the base model.

cs.SE↗
arXiv2026

MockingBird: Context-aware Repository-level Code Translation with Isolated Validation

Despite advances in repository-level code translation, automated type resolution across differing type systems and translation validation (functional-equivalence checking) remain challenging. This paper proposes MockingBird, combining context-aware type resolution with mock-based isolated validation. MockingBird retrieves API documentation and contextual code information per source-language type, prompting an LLM to resolve precise cross-language type mappings, reducing the manual effort of existing approaches. Using this automatically constructed type map, MockingBird employs custom serialization/deserialization to generate equivalent target-language mock objects, validating each translation unit in isolation without test translation, eliminating test-overfitting risks and translation challenges. On 10 Java-to-Python tasks from the AlphaTrans benchmark, MockingBird improves functional equivalence from the state-of-the-art 25.14% to 43.10%, with no manual type-mapping correction needed. A 700-fragment study shows MockingBird achieves 0 false positives and 0.9% false negatives, substantially outperforming GraalVM (2.4% false positives, 7.6% false negatives) and MatchFixAgent (11.0% combined mislabel rate). The pipeline is also LLM-agnostic, maintaining these gains with both DeepSeekCoder-33B and the smaller Devstral-24B.

cs.SE↗
arXiv2026

Learning from Repair Experience: Self-Evolving Agents for Automated Vulnerability Repair

Large Language Models (LLMs) have shown promise for automated vulnerability repair (AVR), but they still face several limitations, including the lack of intra-vulnerability experience accumulation and the lack of cross-vulnerability experience reuse. As a result, LLMs may repeatedly make similar mistakes during iterative repair and underutilize valuable repair knowledge from historical vulnerabilities. To address these challenges, we propose EvoRepair, an experience-based self-evolving AVR agent framework that enables LLMs to accumulate, refine, and leverage domain-specific knowledge across long-horizon vulnerability repairs. EvoRepair follows a cyclic learn-and-repair process that retrieves relevant past experiences to guide repair, extracts new experiences from repair trajectories, and updates an experience bank using quality-aware scoring. We evaluate EvoRepair against 15 representative vulnerability repair baselines on PATCHEVAL and SEC-bench. Results show that EvoRepair achieves the best overall performance, reaching 93.47% on PATCHEVAL, 87.00% on SEC-bench, and 90.46% overall. In particular, EvoRepair outperforms the latest LLM-based baseline LoopRepair by 39.56% and 33.50% on PATCHEVAL and SEC-bench, respectively, and surpasses IntentFix by 70.86% and 50.50%. Across both benchmarks, EvoRepair also exceeds the recent self-evolving agent Live-SWE-Agent by 6.98% overall. Additional transfer experiments on VUL4J further demonstrate the robustness of EvoRepair across models, programming languages, and datasets. These findings demonstrate that experience-based self-evolution substantially strengthens agentic AVR and goes beyond existing self-evolving techniques.

cs.SE↗