Comparative experiments conducted on four checkpoints from a single model family validate a concise and effective scaling trend: end-to-end detection correctness exhibits a complexity-differentiated scaling pattern, and the complexity metric derived from the constraint dependency graph can effectively quantify instance difficulty and the performance improvement potential of models.
Abstract
The increasing complexity of enterprise business scenarios has promoted the widespread adoption of long SKILL documents in agent systems, posing new challenges for compliance detection: large models incur substantial inference costs, while small models may fail to maintain detection accuracy. To address this gap, we propose SkillCDG, a graph-based framework for long SKILL compliance detection. SkillCDG represents complex business policies as a two-layer constraint dependency graph, where the upper layer indexes SKILL descriptions for scenario routing and the lower layer captures dependencies among atomic constraints within each SKILL. During inference, two-level retrieval followed by dependency closure supports compliance judgment and source traceability. We comprehensively evaluate the framework on three enterprise datasets and two controlled public benchmark variants. Experimental results demonstrate that SkillCDG outperforms baseline methods by up to 12.8 percentage points in detection F1 score, while reducing token consumption by a maximum 64.3\%. Moreover, we further investigate the inherent relationships among policy-graph complexity, model scale, and detection performance. Comparative experiments conducted on four checkpoints from a single model family validate a concise and effective scaling trend: end-to-end detection correctness exhibits a complexity-differentiated scaling pattern, and the complexity metric derived from the constraint dependency graph can effectively quantify instance difficulty and the performance improvement potential of models. Leveraging this insightful scaling trend, we conduct adaptive training sample selection and adopt on-policy distillation to efficiently enhance the compliance detection capability of small-scale models.
The paper contributes three reusable deployment patterns: hybrid RAG evidence construction, multi-channel retrieval and reranking produce auditable FAQ candidates, and trace-driven RAG and reranker improvement, where reranker fine-tuning is evaluated not only for in-domain gain but also for forgetting risk.
Long-horizon web agents frequently fail without knowing where or why execution broke down. This issue is particularly pronounced in skill-based agentic web systems, where failures arise within composite skills whose internal decision processes are not directly traceable, making precise diagnosis and repair especially difficult over long horizons. We introduce SkillTracer, a framework that represents skills as attributed plan graphs structured by hierarchical nodes and verifiable edge transitions, enabling programmatic verification of execution progress. By decomposing skills into inspectable hierarchies, SkillTracer converts raw interaction traces into structural evidence, making execution breakdowns localizable to specific node-level decision points and attributable to failing components. This attribution signal facilitates targeted structural repair, allowing the agent to selectively revise failing components while preserving the integrity of valid substructures for partial reuse and adaptive recovery. Furthermore, SkillTracer synthesizes short-term traces with long-term historical evidence to construct a persistent skill graph, enabling failure patterns to drive continual refinement across episodes. Evaluated on challenging long-horizon benchmarks, SkillTracer achieves a 17.7% average improvement in success rate over strong baselines, with gains of up to 56.3% in cross-domain settings, demonstrating that structural attribution and skill repair are critical for reliable long-horizon web interaction. A project page is available at: https://liyuuuuy.github.io/SkillTracer/.
Yuyang Li, Yiran Dou, Jiejing Shao et al.· Proceedings of the 32nd ACM...· 5 citations
SkillZip is proposed, an execution-aware procedural abstraction framework that performs contract-preserving compression over section-level graphs that hydrates a compact, dependency-closed context and expands macros only when required.
Xingyu Tan, Xiaoyang Wang, Qing Liu et al.· 2 citations
Reusable agent skills extend large language model (LLM) agents with task procedures, tool-use guidance, and output constraints. Yet these skills also act as externalized behavioral policies, which create a supply-chain risk: a third-party skill may preserve the declared task and valid output interface while covertly redirecting agent decisions toward an undisclosed objective. We formalize Skill Policy Integrity, which requires a Skill-induced policy to remain aligned with its declared functionality and the user-authorized objective. We further present SkillShift, a constrained black-box framework for covert policy steering without explicit target command injection or task hijacking. It combines semantically plausible policy edits with hierarchical validation, failure-guided optimization, and strategy compression to preserve effectiveness, output validity, transferability, and inconspicuousness. We instantiate this threat in agentic commerce and software dependency use, with SkillShift achieving attacker-favored selection rates of 81.33% and 63.33% while maintaining a 100% utility-preserving rate. The frozen policies also transfer without further optimization across heterogeneous LLM backends and agent environments. Moreover, the evaluated scanners fail to detect the constructed skills, motivating behavioral auditing of reusable skills as agent policy artifacts.
Jia-Rui Li, Jiahao Chen, Chunyi Zhou et al.· 0 citations
TKFQA, a factuality consistency benchmark comprising 10,130 question-answering (QA) pairs grounded in tables, texts, and knowledge graphs, is introduced and ORLF, an LLM-agnostic training framework that models cross-context topological relations through knowledge-specific latent vectors is proposed.
Shibo Chu, Yuze Liu, Tiehua Zhang et al.· 0 citations
This work proposes a framework that ensembles the reasoning structure, not just the answers, of multiple LLMs by weighted merging of Directed Acyclic Graphs (DAGs) extracted from reasoning chains by weighted merging of Directed Acyclic Graphs (DAGs) extracted from reasoning chains.
Amruta Parulekar, Jinu Lee, Dilek Z. Hakkani-Tür et al.· arXiv.org· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
