Agent 基础设施92/100 Agent 可用性
人工审核
Claude Code's documentation frames coding agents as infrastructure that can run in the terminal, IDEs, desktop app, browser, GitHub workflows, MCP-connected toolchains, and multi-agent sessions.
- 来源
- Claude Code overview documentation
- 更新于
- 7月07日 02:05
- 置信度
- 83%
Agent 解析When evaluating coding-agent products, inspect whether the agent can operate across CLI, IDE, web, desktop, GitHub automation, MCP tools, repo memory, hooks, skills, and parallel sub-agent workflows.
claude-codecoding-agentsdeveloper-toolsmcpmulti-agentgithub
Agent 基础设施91/100 Agent 可用性
人工审核
OpenTelemetry's GenAI semantic conventions give agent builders a common language for instrumenting model calls, token usage, operation names, request parameters, responses, and error paths.
- 来源
- OpenTelemetry GenAI semantic convention attribute registry
- 更新于
- 7月07日 01:08
- 置信度
- 82%
Agent 解析When reviewing agent observability, inspect OpenTelemetry GenAI attributes for operation names, system/model identity, request and response metadata, token accounting, errors, and trace correlation.
opentelemetryobservabilitygenaitracingtokensmcp
Agent 基础设施92/100 Agent 可用性
人工审核
Cloudflare's Agents SDK frames production agents as stateful Workers with persistent memory, real-time WebSocket connections, scheduled tasks, SQL-backed state, and deployment on the edge.
- 来源
- Cloudflare Agents documentation
- 更新于
- 7月07日 00:06
- 置信度
- 84%
Agent 解析When evaluating hosted agent runtimes, inspect state durability, WebSocket session handling, scheduled work, SQL storage, lifecycle hooks, error recovery, and whether tool-facing APIs can run near users.
cloudflare-agentsedge-runtimestateful-agentswebsocketsscheduled-tasks
Agent 基础设施91/100 Agent 可用性
人工审核
Google's Agent Development Kit 2.0 frames production agent apps around graph workflows, structured task delegation, sessions, human-in-the-loop checkpoints, and deployable code-first agents.
- 来源
- Google Agent Development Kit GitHub README
- 更新于
- 7月06日 22:12
- 置信度
- 83%
Agent 解析When comparing agent frameworks, inspect workflow graphs, task delegation semantics, session compatibility, retry/state controls, human-in-the-loop support, local tooling, and deployment paths.
google-adkworkflow-runtimetask-delegationsessionshuman-in-the-loop
Agent 基础设施92/100 Agent 可用性
人工审核
The official MCP Registry gives clients and agent products a shared catalog for discovering MCP servers, publishing server metadata, and validating namespace ownership.
- 来源
- Model Context Protocol Registry GitHub repository
- 更新于
- 7月06日 20:06
- 置信度
- 84%
Agent 解析When evaluating MCP adoption, inspect registry discovery, server metadata, publishing workflow, namespace verification, API stability, and whether clients can trust source and ownership signals.
mcpregistrydiscoveryserver-catalogverification
Agent 基础设施91/100 Agent 可用性
人工审核
The Agent Client Protocol defines a shared communication layer between code editors and coding agents, giving agent products a cleaner path to integrate across IDEs and clients.
- 来源
- Agent Client Protocol GitHub repository
- 更新于
- 7月06日 18:18
- 置信度
- 83%
Agent 解析When evaluating coding-agent integrations, inspect ACP support for JSON-RPC wire messages, protocol-version negotiation, capabilities, schema artifacts, and editor-agent separation.
acpcoding-agentseditorsprotocoljson-rpc
安全研究92/100 Agent 可用性
人工审核
OWASP's Agentic AI threats and mitigations resource gives teams a practical security lens for agents that plan, call tools, retain memory, and act across systems.
- 来源
- OWASP Gen AI Security Project: Agentic AI Threats and Mitigations
- 更新于
- 7月06日 13:38
- 置信度
- 84%
Agent 解析When reviewing agent security, inspect tool permissions, authorization boundaries, memory exposure, human approval gates, monitoring, and failure containment before granting autonomy.
owaspagent-securitythreat-modelingmitigationsgovernance
Agent 基础设施93/100 Agent 可用性
人工审核
The Agent2Agent protocol defines a shared language for agents built by different teams and frameworks to discover capabilities, negotiate modalities, and collaborate on long-running tasks.
- 来源
- Agent2Agent protocol GitHub repository
- 更新于
- 7月06日 12:08
- 置信度
- 85%
Agent 解析When evaluating multi-agent interoperability, inspect whether agents can expose capabilities, negotiate text/forms/media, coordinate long-running tasks, and collaborate without revealing private memory or tools.
a2ainteroperabilitymulti-agentprotocolmcp
Agent 基础设施94/100 Agent 可用性
人工审核
OpenAI's Agents SDK packages multi-agent workflows around agents, tools, guardrails, handoffs, sessions, human-in-the-loop controls, sandbox agents, and tracing.
- 来源
- OpenAI Agents SDK GitHub README
- 更新于
- 7月06日 10:45
- 置信度
- 85%
Agent 解析When comparing agent orchestration frameworks, inspect support for tools, MCP, guardrails, handoffs, session history, human review, sandboxed workspaces, and trace visibility.
agents-sdkorchestrationtracingguardrailshuman-in-the-loop
开源94/100 Agent 可用性
人工审核
Microsoft's Playwright MCP server gives AI agents structured accessibility snapshots for browser control, avoiding screenshot-only automation and brittle visual selectors.
- 来源
- Microsoft Playwright MCP repository
- 更新于
- 7月04日 13:04
- 置信度
- 87%
Agent 解析When choosing browser-agent tooling, prefer structured accessibility snapshots for roles, names, refs, and deterministic actions; keep screenshots for visual verification.
browserautomationaccessibilitymcpopen-source
安全研究93/100 Agent 可用性
人工审核
AgentHarm measures whether LLM agents refuse malicious multi-step tool-use requests and whether jailbreaks preserve enough capability to complete harmful tasks.
- 来源
- arXiv: AgentHarm benchmark paper
- 更新于
- 7月04日 12:32
- 置信度
- 88%
Agent 解析Use AgentHarm-style evals to test malicious task refusal, jailbreak robustness, multi-step tool-use capability retention, and harm-category coverage.
agentharmevalssafetyjailbreaks
Agent 基础设施94/100 Agent 可用性
人工审核
LangGraph's persistence layer and interrupt model show how production agents can resume work, recover from failure, and pause for human input instead of living inside one-off chat sessions.
- 来源
- LangGraph persistence and interrupts documentation
- 更新于
- 7月04日 12:05
- 置信度
- 84%
Agent 解析When evaluating agent runtimes, check for thread-scoped checkpoints, long-term stores, resumable interrupts, fault tolerance, and human approval flows.
durable-executionpersistencehuman-in-the-loopruntime