Tag
The paper formulates Task-Conditioned Agent Resource Composition (AgentComposition), proposing a retrieval-grounded, bundle-aware generation method that jointly selects a backbone model and composes a tool–skill bundle, evaluated on a new AgentSelect-Exec benchmark of 5,060 executed configurations with strong gains over baselines.
This paper studies how small an encoder can be for closed-set decisions like tool selection, finding that per-option scoring beats joint scoring and that 68M/150M parameter joint encoders outperform a 435M DeBERTa-v3 zero-shot classifier while answering requests in milliseconds.
Ogcode 的团队分享了其用户数据洞察:开发者安装大量 skills 和 MCP 工具导致 agent 选错工具并浪费高达 77% 的 token,他们正在研发能按需动态选择工具的解决方案,并向社区征集思路。
论文分析了 367,174 个 MCP 工具的元数据,通过受控干预实验发现参数 schema 是影响 LLM 工具选择的最强因素,超过名称和描述。研究指出 schema 设计与工具排序会影响模型调用哪个实现,即使功能相同。
This paper presents a paired, self-audited evaluation of two System-1 decision models (open-weight Laya and hosted Jev) across 11 agent-harness decision points, finding Jev significantly more accurate on 9 of 11 tasks while auditing several analysis errors that had distorted prior deployment claims.
ToolSearcher is a reinforcement learning framework designed to optimize tool selection for large language models in large-scale tool repositories, using techniques such as category-constrained discrimination and event-level search modeling to improve performance in iterative search and tool composition tasks.
Harness Router is a decision layer for AI coding agents that improves tool selection before execution using Jev for ambiguity and MCTS for multi-step consequences.
The article discusses design patterns for integrating Jev with LLMs in AI agents, specifically how agents handle tool selection and planning when only one tool schema is exposed at a time.
Toollery is a training-free candidate-compression framework that improves scalability and efficiency for LLM agent tool and skill selection through retrieval-based methods.
The author updates on a read-only analytics agent project using Gemini Flash Lite, discusses caching inefficiencies and bilingual routing challenges, and seeks advice on date handling.
The paper introduces Gavel, a method that elicits native skill routing from frozen LLMs via linear projections, enabling efficient tool selection without context overload and outperforming existing pipelines on benchmarks.
A developer describes building a read-only analytics agent for restaurant POS systems using Node.js, TypeScript, and Gemini Flash, seeking advice on tool selection, grounding techniques, and patterns for implementing write actions.
A study measured 16,893 sessions to analyze how AI coding agents like Claude Code, Codex, and Cursor select tools such as databases, finding consistent recommendations across varied contexts.
This article offers a practical framework for evaluating AI agent tools by focusing on key questions about job clarity, access needs, autonomy, maintenance, cost, and exit options, moving beyond feature comparisons.
The author observed that adding more tools to an AI agent decreased its accuracy in selecting the correct tool due to increased classification complexity, and found that using multiple smaller agents with narrower toolsets improved reliability.
This paper presents a case study on skill discovery and routing in a multimodal agent harness, showing that partial in-prompt exposure of skills can create lexical competition that hinders correct selection, linking small-scale in-context retrieval to large-scale approaches.
OpenMed demonstrated using Liquid AI's LFM2.5-VL-3B model to analyze a generated skin image, mapping regions and measuring dimensions locally on a Mac Studio for visual review.
Unused tools in an AI agent's toolset still consume tokens and add noise to tool selection, so agents should load only the tools required for the current task.
This paper introduces 'canary tools' as diagnostic probes to identify specific tool-selection reasoning failures in LLM agents, proposing a six-type taxonomy and evaluating eight models across 8,640 task runs to show capability-tier disparities and robustness.
Ratel is an open-source tool that reduces input tokens by 79% and improves tool selection accuracy for AI agents by loading only needed tools using a BM25 index, instead of all available tools.