Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Unverified50% confidenceSolutionExact time

工具描述的精确程度直接决定Agent的任务执行成功率，模糊或歧义的描述是导致工具误调用最常见的根本原因

1

Sources

50%

Confidence

Long-term

Relevance

7/14/2026

First Seen

Sources

AI Agent开发实战：框架选型、工具调用与落地部署完整指南

Related Claims

Unverified语义层与执行层的割裂是业务Skill失效的根本原因，语义层描述模糊会导致Agent从不调用或错误场景频繁触发该Skill79% similar Unverified工具设计应遵循动作幂等、错误明确的原则，否则模型调用失败无法恢复，工具描述质量直接决定Agent成败77% similar Unverified模糊的Skill描述会导致模型错误认为某Skill适用于当前场景，这一现象被称为工具幻觉（Tool Hallucination）77% similar Unverified在 Skill 或全局配置中提供正例和反例的对比示例（few-shot contrastive examples）能有效纠正 Agent 反复出现的错误74% similar UnverifiedAgent产品在边界情况下（如需求表述模糊、多工具调用失败传递）的准确性与容错能力是当前所有Agent产品的共同技术挑战74% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/507723

API

curl https://kongchang.com/api/v1/knowledge/claims/507723

MCP

get_claim(id=507723)