跳到正文- Harness Engineering: The Work Around the Model
- The Anatomy of a Modern AI Agent
- Web Search APIs for Agents
- Voice Agents
- Verification and evals
- Vercel AI SDK
- Tools
- Subagents
- Source: τ²-bench: Evaluating Conversational Agents in a Dual-Control Environment (Barres et al., 2025)
- Source: τ-bench: Tool-Agent-User Interaction Benchmark (Yao et al., 2024)
- Source: Writing effective tools for agents (Anthropic, 2025)
- Source: Workflow SDK docs: workflows, steps, hooks and webhooks (Vercel)
- Source: Workflow concepts and orchestrations (Microsoft Agent Framework docs)
- Source: Work with sessions (Claude Agent SDK docs)
- Source: Why we no longer evaluate SWE-bench Verified (OpenAI, 2026)
- Source: Why we built ADK 2.0 (Google Developers Blog)
- Source: Why Do Multi-Agent LLM Systems Fail? (Cemri et al., 2025)
- Source: Welcome to ADK 2.0 (ADK docs)
- Source: Wei et al. 2025: BrowseComp
- Source: Voice AI & Voice Agents: An Illustrated Primer
- Source: Vercel Sandbox docs (overview, security model, pricing and quotas)
- Source: Using Agent Skills with the API (Claude Platform docs)
- Source: Tracing (OpenAI Agents SDK docs)
- Source: Tools, agents as tools (OpenAI Agents SDK docs)
- Source: The lethal trifecta for AI agents
- Source: The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management (Lindenbauer et al., 2025)
- Source: The Anatomy of an Agent Harness (LangChain, 2026)
- Source: The Anatomy of a Modern AI Agent
- Source: Terminal-Bench (Merrill et al., 2026)
- Source: Temporal OpenAI Agents SDK integration (Python docs)
- Source: Temporal docs: Workflow definitions, Event History, Activities and retry policies
- Source: Tavily API docs (Search endpoint and credits)
- Source: Tasks (CrewAI docs)
- Source: SWE-bench: Can Language Models Resolve Real-World GitHub Issues? (Jimenez et al., 2023)
- Source: SWE-Bench Pro (Deng et al., Scale AI, 2025)
- Source: SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering (Yang et al., 2024)
- Source: Subagents in the SDK (Claude Agent SDK docs)
- Source: State (ADK sessions docs)
- Source: Spider 2.0 (Lei et al., ICLR 2025) and Leaderboard
- Source: smolagents on PyPI
- Source: smolagents docs: Tools (advanced usage)
- Source: smolagents docs: Secure code execution
- Source: smolagents docs: Manage your agent's memory
- Source: smolagents docs: Inspecting runs with OpenTelemetry
- Source: smolagents docs: Index and Guided tour (Hugging Face)
- Source: smolagents conceptual guides: What are agents? and How do multi-step agents work?
- Source: smolagents API reference: Agents (v1.26.0)
- Source: Sleep-time Compute: Beyond Inference Scaling at Test-time (Lin et al., 2025)
- Source: Sleep-time Compute (Letta blog, 2025)
- Source: SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks (Li et al., 2026)
- Source: Skill authoring best practices (Claude Platform docs)
- Source: Shell tool guide (OpenAI docs)
- Source: Shao et al. 2024: STORM
- Source: Sessions (OpenAI Agents SDK docs)
- Source: SerpApi Google Search API docs
- Source: Semantic Kernel to Microsoft Agent Framework Migration Guide (Microsoft Learn)
- Source: Running agents (OpenAI Agents SDK docs)
- Source: ReWOO: Decoupling Reasoning from Observations (Xu et al., 2023)
- Source: Restate docs: durable execution and the human-in-the-loop agent pattern
- Source: Reflexion: Language Agents with Verbal Reinforcement Learning (Shinn et al., 2023)
- Source: Rearchitecting Letta's Agent Loop (Letta, 2025)
- Source: ReAct: Synergizing Reasoning and Acting in Language Models (Yao et al., 2022)
- Source: Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet (Anthropic, 2025)
- Source: Pydantic AI framework comparison pages (vs Vercel AI SDK, vs LangChain/LangGraph)
- Source: Pydantic AI documentation
- Source: Processes (CrewAI docs)
- Source: Plugin manifest reference (Claude Code docs)
- Source: Perplexity API docs (Search API, Agent API models and effort levels)
- Source: OWASP Top 10 for Agentic Applications (2026)
- Source: Orchestrating Agents: Routines and Handoffs (OpenAI Cookbook)
- Source: openai-agents-python GitHub README
- Source: OpenAI Voice Agents Guide
- Source: OpenAI Launches New API, SDK, and Tools to Develop Custom Agents (InfoQ)
- Source: OpenAI deprecations page (search and deep research models)
- Source: OpenAI Deep Research System Card
- Source: OpenAI API pricing, containers row
- Source: OpenAI Agents SDK handoff runtime (turn_resolution.py)
- Source: OpenAI Agents SDK documentation (overview)
- Source: Open-source DeepResearch (Hugging Face blog)
- Source: On the Brittle Foundations of ReAct Prompting for Agentic LLMs (Verma et al., 2024)
- Source: Notebook: Concurrency design
- Source: Notebook experiment 25: Memory Agent
- Source: Notebook experiment 11: Human-in-the-loop
- Source: New prompt injection papers: Agents Rule of Two and The Attacker Moves Second (Willison)
- Source: NeuralTrust, OWASP Top 10 for Agentic Applications 2026
- Source: Modal Sandboxes guide and networking docs
- Source: MINJA, memory injection attacks on LLM agents (Dong et al., 2025)
- Source: microsoft/semantic-kernel README (GitHub)
- Source: microsoft/autogen GitHub README
- Source: microsoft/agent-framework README (GitHub)
- Source: Microsoft Foundry — Agent Service Overview
- Source: Microsoft Agent Framework Version 1.0 (Microsoft DevBlogs)
- Source: Microsoft Agent Framework overview (Microsoft Learn)
- Source: Memory tool (Claude API docs)
- Source: Memory (CrewAI docs)
- Source: MemGPT: Towards LLMs as Operating Systems (Packer et al., 2023)
- Source: Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory (Chhikara et al., 2025)
- Source: Mem0, the token-efficient memory algorithm (Mem0, 2026)
- Source: Mem0 pricing page
- Source: Mem0 documentation, open source overview and Platform vs OSS
- Source: Mem0 documentation, memory concepts
- Source: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (Becker et al., 2025)
- Source: Measuring AI agent autonomy in practice (Anthropic, 2026)
- Source: mastra-ai/mastra repository and licence (GitHub, npm)
- Source: Mastra docs: Workflows overview
- Source: Mastra docs: Supervisor agents
- Source: Mastra docs: Processors
- Source: Mastra docs: Memory overview
- Source: Mastra docs: Getting started, Agents overview and Agent reference
- Source: Mastra docs: Evals (scorers) overview
- Source: Making Claude Code more secure and autonomous with sandboxing (Anthropic, 2025)
- Source: Loop Engineering (Loic's Notebook)
- Source: LongMemEval (Wu et al., 2024)
- Source: LlamaIndex Workflows documentation
- Source: LiveKit, Using a Transformer to Improve End of Turn Detection
- Source: Lifecycle hooks reference (OpenAI Agents SDK, Python)
- Source: Lies, Damn Lies, & Statistics: Is Mem0 Really SOTA in Agent Memory? (Zep, 2025)
- Source: Letta documentation, self-hosting with Docker (deprecated)
- Source: Letta documentation, memory blocks and archival memory
- Source: Letta documentation, MemFS, memory and dreaming, platform overview
- Source: Language Agent Tree Search (LATS, Zhou et al., 2023)
- Source: LangGraph memory documentation
- Source: LangChain multi-agent docs (overview and handoffs)
- Source: LangChain agents, middleware and structured output docs
- Source: Jina Reader (r.jina.ai / s.jina.ai)
- Source: Introduction to the Deep Research API (OpenAI Cookbook)
- Source: Introducing SWE-bench Verified (OpenAI, 2024)
- Source: Introducing smolagents (Hugging Face blog)
- Source: Intercept and control agent behavior with hooks (Claude Agent SDK docs)
- Source: Inspect documentation (UK AI Security Institute)
- Source: Inngest docs: how functions are executed, step.waitForEvent and usage limits
- Source: Improving agent with semantic search (Cursor, 2025)
- Source: Idempotent requests (Stripe API reference)
- Source: Human-in-the-loop middleware (LangChain docs)
- Source: Human-in-the-loop (OpenAI Agents SDK docs)
- Source: huggingface/smolagents README (GitHub)
- Source: Huang et al. 2025: Deep Research Agents survey
- Source: How we built our multi-agent research system (Anthropic)
- Source: How the agent loop works (Claude Agent SDK docs)
- Source: How Claude remembers your project (Claude Code memory docs)
- Source: How Claude Code works (Anthropic docs)
- Source: Hosting the Agent SDK (Claude Code docs)
- Source: Hooks reference (Claude Code docs)
- Source: Hooks (OpenAI Codex docs)
- Source: Hooks (Gemini CLI docs)
- Source: Holistic Agent Leaderboard (Kapoor et al., 2025)
- Source: Harness vs Loop vs Graph Engineer (Notebook tutorial)
- Source: Harness Engineering: The Work Around the Model
- Source: Harness engineering: leveraging Codex in an agent-first world (OpenAI, 2026)
- Source: Harness design for long-running application development (Anthropic, 2026)
- Source: Handoffs (OpenAI Agents SDK docs)
- Source: Handoff orchestration (Microsoft Agent Framework docs)
- Source: gVisor documentation (overview and platforms)
- Source: Guardrails (OpenAI Agents SDK docs)
- Source: Give Claude custom tools (Claude Agent SDK docs)
- Source: GitHub Copilot cloud (coding) agent docs: overview and risks
- Source: Generative Agents: Interactive Simulacra of Human Behavior (Park et al., 2023)
- Source: GEMINI.md context files (Gemini CLI docs)
- Source: Gemini Enterprise Agent Platform — Agent Runtime Overview (formerly Vertex AI Agent Engine)
- Source: Gemini Deep Research agent docs
- Source: Gemini API Grounding with Google Search docs
- Source: Gartner predicts 40%+ of agentic AI projects canceled by 2027 (2025)
- Source: Flows (CrewAI docs)
- Source: Firecrawl docs (introduction and scrape)
- Source: Firecracker (project site and NSDI 2020 paper)
- Source: Extensions (Gemini CLI docs)
- Source: Extend Claude with skills (Claude Code docs)
- Source: Extend Claude Code (Claude Code docs)
- Source: Executable Code Actions Elicit Better LLM Agents (CodeAct, Wang et al., 2024)
- Source: Exa search API docs
- Source: Event loop (ADK runtime docs)
- Source: Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents? (Gloaguen et al., 2026)
- Source: Equipping agents for the real world with Agent Skills (Anthropic Engineering)
- Source: Effective harnesses for long-running agents (Anthropic, 2025)
- Source: Effective context engineering for AI agents (Anthropic)
- Source: E2B documentation (sandboxes, internet access, secrets, security)
- Source: Don't Build Multi-Agents (Cognition, 2025)
- Source: Design Patterns for Securing LLM Agents against Prompt Injections
- Source: Demystifying evals for AI agents (Anthropic)
- Source: Daytona documentation (sandboxes and network limits)
- Source: Custom instructions with AGENTS.md (OpenAI Codex docs)
- Source: Custom commands (Gemini CLI docs)
- Source: Cursor Rules (Cursor docs)
- Source: CSA research note, OWASP's 2026 LLM Top 10 and Agent Control Standard
- Source: Crews (CrewAI docs)
- Source: CrewAI introduction (CrewAI docs)
- Source: crewAI GitHub README
- Source: Conversations & Memory overview (Microsoft Agent Framework docs)
- Source: Context Engineering for AI Agents: Lessons from Building Manus (Ji, 2025)
- Source: Context editing (Claude API docs)
- Source: Configure permissions (Claude Agent SDK docs)
- Source: Compaction overview (Claude API docs)
- Source: Compaction on demand (Claude API docs)
- Source: Compaction at a token threshold (Claude API docs)
- Source: Collaborative workflows (ADK docs)
- Source: Cognitive Architectures for Language Agents (Sumers et al., 2023)
- Source: Codex agent approvals and security (OpenAI docs)
- Source: CodeAct (Microsoft Agent Framework docs)
- Source: Code Interpreter tool guide (OpenAI docs)
- Source: Code execution with MCP (Anthropic, 2025)
- Source: Code execution tool (Anthropic docs)
- Source: Code execution (Gemini API docs)
- Source: Claude web search tool docs
- Source: Claude web fetch tool docs
- Source: Choose a design pattern for your agentic AI system (Google Cloud)
- Source: CHESS (Talaei et al., 2024)
- Source: Building effective agents (Anthropic)
- Source: Building durable AI agents (Vercel knowledge base guide)
- Source: Building agents with the Claude Agent SDK (Anthropic)
- Source: Build skills (OpenAI Codex docs)
- Source: Brave Search API product page
- Source: BFCL V3: Multi-Turn and Multi-Step Function Calling (Berkeley)
- Source: Best practices for Claude Code (Anthropic docs)
- Source: Benchmarking AI Agent Memory: Is a Filesystem All You Need? (Letta, 2025)
- Source: AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation (Wu et al., 2023)
- Source: AutoGen vs CrewAI (Loic's Notebook)
- Source: AutoGen v0.4: Reimagining the foundation of agentic AI (Microsoft Research)
- Source: AutoGen to Microsoft Agent Framework Migration Guide (Microsoft Learn)
- Source: ARC-AGI series and ARC-AGI-3 (ARC Prize Foundation)
- Source: Announcing Mastra 1.0 (Mastra blog)
- Source: Amazon Bedrock AgentCore — Overview
- Source: Amazon Bedrock AgentCore — How Runtime Works
- Source: Amazon Bedrock — Agents (Classic)
- Source: AI SDK documentation (Vercel)
- Source: AI SDK 7 is now available (Vercel changelog)
- Source: AI Agents That Matter (Kapoor et al., 2024)
- Source: AGENTS.md (open format site)
- Source: Agents Rule of Two (Meta)
- Source: Agents (OpenAI Agents SDK docs)
- Source: Agents (CrewAI docs)
- Source: agentevals (LangChain)
- Source: Agent Skills specification (agentskills.io)
- Source: Agent Skills overview (Claude Platform docs)
- Source: Agent Skills (Gemini CLI docs)
- Source: Agent Skills (Cursor docs)
- Source: Agent SDK overview (Claude Code docs)
- Source: Agent Development Kits (Loic's Notebook)
- Source: Agent Development Kit (ADK) documentation home
- Source: Agent concepts and Agent Pipeline Architecture (Microsoft Agent Framework docs)
- Source: ADK workflows, graph workflows and LoopAgent (ADK docs)
- Source: ADK documentation: Why evaluate agents
- Source: ADK agent transfer (adk-python source and LlmAgent docs)
- Source: A2A Protocol Specification v1.0
- Source: A2A in ADK (ADK docs)
- Source: A sandbox without a network boundary is only half a sandbox (Vercel, 2026)
- Source: A Practical Guide to Building Agents (OpenAI)
- smolagents
- ReAct pattern
- Pydantic AI
- Prompt injection
- OWASP Top 10 for LLM and Agentic Applications
- OpenTelemetry GenAI Semantic Conventions
- OpenAI Agents SDK
- Model Context Protocol (MCP)
- Microsoft Agent Framework
- Mem0
- Mastra
- LLM security risks
- LLM Red Teaming
- LLM Evaluation
- LlamaIndex
- Letta
- Lethal trifecta
- LangGraph
- LangChain
- Human-in-the-Loop
- Harness Engineering
- Google Vertex AI
- Google ADK
- Error Analysis
- Durable Agent Execution
- Deep Research Agents
- CrewAI
- Context engineering
- Context Compaction
- Computer-Use Agents
- Coding Agents
- Code Execution Sandboxes
- Claude Agent SDK
- Azure AI Foundry
- AutoGen and Microsoft Agent Framework
- Amazon Bedrock
- Agentic RAG
- Agent2Agent (A2A) Protocol
- Agent Skills
- Agent Memory
- Agent loop
- Agent Instruction Files
- Agent Hooks and Middleware
- Agent Handoffs
- Agent guardrails
- Agent Evaluation
- Agent design patterns
- Agent Authentication and Authorization