1
OpenAI Agents API 开放公测：支持代码执行、工具调用和跨上下文任务运行，为开发者提供云端智能体基础设施
IT之家 9 月 11 日消息，OpenAI 于当地时间 9 月 10 日宣布推出 Agents API 公测版，允许开发者通过 API 调用由 OpenAI 管理的云端 AI 智能体运行环境。 该服务复用了 Codex 背后的智能体执行框架和基础设施，开发者可以指定任务、模型、工具及运行环境，打造…
2
LLM Tool Failures: Only 3 Root Causes – Value, Condition, Intent
大模型工具调用总翻车？三大根因一次讲透，从“填表人”视角避开90%的坑。
Article URL: https://github.com/Jang-woo-AnnaSoft/execution-state-preflight/blob/main/who-fills-in-the-form.md Comments URL: https://news.ycombinator.…
3
Runtime Governance for Agentic AI: Action-Boundary Control with Trusted Provenance and Fail-Closed Execution
给自主智能体戴上“安全紧箍”，行动边界+可信溯源+失败即停，杜绝越权操作。
arXiv:2608.16891v1 Announce Type: new Abstract: Agentic AI systems request tool actions that can modify files, send messages, launch jobs, or change w…
4
Building Custom MCP Servers: Extending AI with Tools
讲解如何基于JSON-RPC 2.0搭建自定义MCP服务器，让AI模型按需调用工具，告别繁琐定制集成。
The Protocol That Actually Standardizes Tooling MCP (Model Context Protocol) is essentially a JSON-RPC 2.0 service with a schema for capabilities, too…
5
Show HN: Tracelint – a linter for AI agent traces, no LLM judge
给AI智能体轨迹做“结构体检”，无需LLM裁判，精准揪出工具调用中的模式违规和错误复用。
Article URL: https://github.com/AshwinUgale/tracelint Comments URL: https://news.ycombinator.com/item?id=49346452 Points: 3 # Comments: 0
6
Show HN: Doberman: The AI watchdog that stops Claude from deleting your database
AI安全新防线：Doberman双重监控LLM输入输出与工具调用，阻止Claude等误删数据库，值得一试
Landing Page: https://www.trydoberman.dev/ The Problem: AI guardrails only protect against 89% of dangerous commands. This may seem pretty good until …
7
Tools as Continuous Flow for Evolving Agentic Reasoning
把工具调用当作连续流来驱动智能体推理进化，为下一代Agent架构提供全新视角
arXiv:2605.07339v2 Announce Type: replace Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in orchestrating tools for …
8
Leanroute Is Live One AI Gateway for Models and Tools
AI网关进化新方向，统一调度大模型与MCP工具流量，为AI应用提供一体化接入层。
Hi all, We built Leanroute - it's One AI Gateway for LLM and Tools. Market is flooded with LLM Gateways, so I wanted to build an AI Gateway for LLM an…
9
Prompts Don't Protect: Architectural Enforcement via MCP Proxy for LLM Tool Access Control
AI代理工具调用存在致命漏洞，MCP代理强制访问控制补齐安全短板，值得关注。
arXiv:2605.18414v2 Announce Type: replace-cross Abstract: Large language models increasingly operate as autonomous agents that select and invoke tools…
10
TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories
系统评估大模型多步工具调用中的安全护栏，为Agent可靠性提供关键基准。
arXiv:2604.07223v2 Announce Type: replace-cross Abstract: As large language models (LLMs) evolve from static chatbots into autonomous agents, the prim…
11
The Bitter Lesson of Tool Calling
从工具调用看AI发展的长期主义：通用能力终将胜过精巧设计，值得每位关注智能体的读者深思。
arXiv:2608.06370v1 Announce Type: new Abstract: Tool use transforms LLMs into agents that act beyond their training data, and for code-capable models,…
12
FinToolBench: Evaluating LLM Agents for Real-World Financial Tool Use
金融领域首个评估大模型工具调用能力的基准，看智能体能否玩转真实金融工具
arXiv:2603.08262v2 Announce Type: replace Abstract: The integration of Large Language Models (LLMs) into the financial domain is driving a paradigm sh…
13
Tool Specifications Matter: Uncovering and Mitigating Safety Risks in AI Agents
AI智能体接入外部工具后安全性骤降，这份研究揭示工具规范细节如何诱发风险并给出缓解思路，值得关注
arXiv:2607.29254v1 Announce Type: new Abstract: AI agents extend large language models (LLMs) with external tools, enabling them to perform complex ta…
14
Teaching an AI to know itself: Building a local LLM agent in D
用D语言打造能自我认知的本地LLM代理，揭秘工具调用与自我交互的实现细节。
Article URL: https://blog.dlang.org/2026/06/07/teaching-an-ai-to-know-itself-building-a-local-llm-agent-in-d/ Comments URL: https://news.ycombinator.c…
15
Show HN: GAI – A Go runtime for typed, tool-using LLM agents
面向Go开发者的类型化LLM代理运行时，支持工具调用，代码开源可直接上手。
Article URL: https://github.com/lace-ai/gai Comments URL: https://news.ycombinator.com/item?id=49129430 Points: 1 # Comments: 0
16
Go LLM SDK for streaming, tool-calling AI backends (plus frontend React lib)
Grafana 出品的 Go LLM SDK，支持流式传输与工具调用，还附带 React 前端库，AI 应用开发利器。
Article URL: https://github.com/grafana/ai-sdk Comments URL: https://news.ycombinator.com/item?id=49108778 Points: 46 # Comments: 13
17
Flat Score, Amplified Failures: How the Error Budget Masks Damage in Quantized LLM Agents
4-bit量化看似无损，但工具调用型Agent的多轮任务中错误被悄悄放大，一份扎实的基准评测揭穿幻觉。
arXiv:2607.27275v1 Announce Type: new Abstract: Post-training quantization to 4-bit weights is widely reported to be nearly lossless. We test this cla…
18
Beyond Aggregate Risk: Role-Stratified Conformal Risk Control for LLM Tool Calls
这篇论文提出了“角色分层共形风险控制”新方法，专门应对LLM工具调用中不同角色的差异化风险，而非只关注平均指标。
arXiv:2607.24343v1 Announce Type: new Abstract: Language-model agents act through structured tool calls whose arguments carry different risks. Untrust…
19
Show HN: OpenCode Session Tracker
实时追踪AI代理的思考与工具调用，支持操作审批，开源神器来了！
Article URL: https://github.com/ykocaman/opencode-session-tracker/tree/main Comments URL: https://news.ycombinator.com/item?id=49018453 Points: 1 # Co…
20
Proxy for OpenAI Codex and Claude Code, use any LLM with those apps
一个轻量本地代理，让Codex和Claude Code兼容任意LLM，支持流式、工具调用和推理。
Article URL: https://github.com/lidge-jun/opencodex Comments URL: https://news.ycombinator.com/item?id=49012330 Points: 2 # Comments: 0