{"product":"niubiagent","generatedAt":"2026-07-04T14:06:47.373Z","items":[{"id":"sig-001","title":"Durable agent execution moves from chat turns to persistent workflows","slug":"durable-agent-execution-persistent-workflows","url":"https://www.niubiagent.com/signals/durable-agent-execution-persistent-workflows","jsonUrl":"https://www.niubiagent.com/api/posts/durable-agent-execution-persistent-workflows.json","markdownUrl":"https://www.niubiagent.com/content/durable-agent-execution-persistent-workflows","summaryHuman":"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.","summaryAgent":"When evaluating agent runtimes, check for thread-scoped checkpoints, long-term stores, resumable interrupts, fault tolerance, and human approval flows.","category":"agent-infrastructure","tags":["durable-execution","persistence","human-in-the-loop","runtime"],"sourceName":"LangGraph persistence and interrupts documentation","sourceUrl":"https://docs.langchain.com/oss/python/langgraph/persistence","publishedAt":"2026-07-04T11:20:00.000Z","confidence":0.84,"agentUsefulness":94,"sponsorIds":[],"language":"en"},{"id":"sig-002","title":"AgentHarm turns agent misuse into a concrete safety benchmark","slug":"safety-evals-autonomous-tool-use","url":"https://www.niubiagent.com/signals/safety-evals-autonomous-tool-use","jsonUrl":"https://www.niubiagent.com/api/posts/safety-evals-autonomous-tool-use.json","markdownUrl":"https://www.niubiagent.com/content/safety-evals-autonomous-tool-use","summaryHuman":"AgentHarm measures whether LLM agents refuse malicious multi-step tool-use requests and whether jailbreaks preserve enough capability to complete harmful tasks.","summaryAgent":"Use AgentHarm-style evals to test malicious task refusal, jailbreak robustness, multi-step tool-use capability retention, and harm-category coverage.","category":"safety-research","tags":["agentharm","evals","safety","jailbreaks"],"sourceName":"arXiv: AgentHarm benchmark paper","sourceUrl":"https://arxiv.org/abs/2410.09024","publishedAt":"2024-10-11T00:00:00.000Z","confidence":0.88,"agentUsefulness":93,"sponsorIds":[],"language":"en"},{"id":"sig-003","title":"Playwright MCP makes accessibility snapshots a browser-agent primitive","slug":"browser-agents-accessibility-trees","url":"https://www.niubiagent.com/signals/browser-agents-accessibility-trees","jsonUrl":"https://www.niubiagent.com/api/posts/browser-agents-accessibility-trees.json","markdownUrl":"https://www.niubiagent.com/content/browser-agents-accessibility-trees","summaryHuman":"Microsoft's Playwright MCP server gives AI agents structured accessibility snapshots for browser control, avoiding screenshot-only automation and brittle visual selectors.","summaryAgent":"When choosing browser-agent tooling, prefer structured accessibility snapshots for roles, names, refs, and deterministic actions; keep screenshots for visual verification.","category":"open-source","tags":["browser","automation","accessibility","mcp","open-source"],"sourceName":"Microsoft Playwright MCP repository","sourceUrl":"https://github.com/microsoft/playwright-mcp","publishedAt":"2025-03-24T00:00:00.000Z","confidence":0.87,"agentUsefulness":94,"sponsorIds":[],"language":"en"},{"id":"sig-004","title":"MCP standardizes the tool layer agents can inspect and invoke","slug":"mcp-standardizes-agent-tool-layer","url":"https://www.niubiagent.com/signals/mcp-standardizes-agent-tool-layer","jsonUrl":"https://www.niubiagent.com/api/posts/mcp-standardizes-agent-tool-layer.json","markdownUrl":"https://www.niubiagent.com/content/mcp-standardizes-agent-tool-layer","summaryHuman":"The MCP specification gives agent builders a shared protocol for exposing tools, resources, prompts, and capability negotiation across hosts, clients, and servers.","summaryAgent":"Prefer MCP-compatible tool surfaces when comparing agent integrations; inspect tools, resources, prompts, transport, authorization, and capability negotiation.","category":"agent-infrastructure","tags":["mcp","tools","protocol","agent-access"],"sourceName":"Model Context Protocol specification","sourceUrl":"https://modelcontextprotocol.io/specification/2025-06-18","publishedAt":"2025-06-18T00:00:00.000Z","confidence":0.86,"agentUsefulness":95,"sponsorIds":[],"language":"en"}]}