<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[长任务恢复别靠“继续刚才”]]></title><description><![CDATA[<p dir="auto">Agent 跑 40 分钟的资料整理，中途断了。现在只能让它“继续刚才”，结果它重复做了半天。长任务怎么恢复？</p>
]]></description><link>https://localaihub.com/topic/118/长任务恢复别靠-继续刚才</link><generator>RSS for Node</generator><lastBuildDate>Wed, 03 Jun 2026 20:31:36 GMT</lastBuildDate><atom:link href="https://localaihub.com/topic/118.rss" rel="self" type="application/rss+xml"/><pubDate>Thu, 07 May 2026 23:01:00 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 19:44:00 GMT]]></title><description><![CDATA[<p dir="auto">这就能跑起来。长任务不是靠模型记忆强，是靠工程状态清楚。</p>
]]></description><link>https://localaihub.com/post/1055</link><guid isPermaLink="true">https://localaihub.com/post/1055</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Fri, 08 May 2026 19:44:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 19:12:00 GMT]]></title><description><![CDATA[<p dir="auto">我先拆成阶段：采集、清洗、归类、人工抽检、导入。每阶段写状态和产物。</p>
]]></description><link>https://localaihub.com/post/1054</link><guid isPermaLink="true">https://localaihub.com/post/1054</guid><dc:creator><![CDATA[好困但在看日志]]></dc:creator><pubDate>Fri, 08 May 2026 19:12:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 18:06:00 GMT]]></title><description><![CDATA[<p dir="auto">补偿也要谨慎。删除刚创建的东西可能比重复创建更危险。</p>
]]></description><link>https://localaihub.com/post/1053</link><guid isPermaLink="true">https://localaihub.com/post/1053</guid><dc:creator><![CDATA[小傅]]></dc:creator><pubDate>Fri, 08 May 2026 18:06:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 17:10:00 GMT]]></title><description><![CDATA[<p dir="auto">外部动作加 operation_id。恢复时先查这个 id 是否已完成，再决定跳过或补偿。</p>
]]></description><link>https://localaihub.com/post/1052</link><guid isPermaLink="true">https://localaihub.com/post/1052</guid><dc:creator><![CDATA[nora]]></dc:creator><pubDate>Fri, 08 May 2026 17:10:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 16:11:00 GMT]]></title><description><![CDATA[<p dir="auto">还有幂等性。导入数据、发消息、创建工单这种，恢复时不能重复执行。</p>
]]></description><link>https://localaihub.com/post/1051</link><guid isPermaLink="true">https://localaihub.com/post/1051</guid><dc:creator><![CDATA[Grace]]></dc:creator><pubDate>Fri, 08 May 2026 16:11:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 13:12:00 GMT]]></title><description><![CDATA[<p dir="auto">每步产物最好可校验。比如抓网页，记录 URL、状态码、文本长度、截图路径。别只写“已抓取”。</p>
]]></description><link>https://localaihub.com/post/1050</link><guid isPermaLink="true">https://localaihub.com/post/1050</guid><dc:creator><![CDATA[别问问就是缓存]]></dc:creator><pubDate>Fri, 08 May 2026 13:12:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 11:42:00 GMT]]></title><description><![CDATA[<p dir="auto">日志和状态要分开。日志给人审计，状态给机器恢复。</p>
]]></description><link>https://localaihub.com/post/1049</link><guid isPermaLink="true">https://localaihub.com/post/1049</guid><dc:creator><![CDATA[树莓派烫手]]></dc:creator><pubDate>Fri, 08 May 2026 11:42:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 10:28:00 GMT]]></title><description><![CDATA[<p dir="auto">我们之前把完整日志塞回 prompt，越恢复越糟。失败路径也被它当成有效线索。</p>
]]></description><link>https://localaihub.com/post/1048</link><guid isPermaLink="true">https://localaihub.com/post/1048</guid><dc:creator><![CDATA[小周]]></dc:creator><pubDate>Fri, 08 May 2026 10:28:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 09:45:00 GMT]]></title><description><![CDATA[<p dir="auto">对。恢复不是“把历史全塞回去”，而是重建最小必要上下文。</p>
]]></description><link>https://localaihub.com/post/1047</link><guid isPermaLink="true">https://localaihub.com/post/1047</guid><dc:creator><![CDATA[melo]]></dc:creator><pubDate>Fri, 08 May 2026 09:45:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 06:49:00 GMT]]></title><description><![CDATA[<p dir="auto">不是给模型自由看懂，是程序读状态，决定下次给模型什么上下文。</p>
]]></description><link>https://localaihub.com/post/1046</link><guid isPermaLink="true">https://localaihub.com/post/1046</guid><dc:creator><![CDATA[阿宁]]></dc:creator><pubDate>Fri, 08 May 2026 06:49:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 04:55:00 GMT]]></title><description><![CDATA[<p dir="auto">JSON 够吗？会不会模型看不懂？</p>
]]></description><link>https://localaihub.com/post/1045</link><guid isPermaLink="true">https://localaihub.com/post/1045</guid><dc:creator><![CDATA[陈小舟]]></dc:creator><pubDate>Fri, 08 May 2026 04:55:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 04:32:00 GMT]]></title><description><![CDATA[<p dir="auto">我们自己做得很土：每个阶段落一个 JSON，里面写阶段状态和文件列表。断了从最后一个 completed 阶段开始。</p>
]]></description><link>https://localaihub.com/post/1044</link><guid isPermaLink="true">https://localaihub.com/post/1044</guid><dc:creator><![CDATA[阿宁]]></dc:creator><pubDate>Fri, 08 May 2026 04:32:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 03:15:00 GMT]]></title><description><![CDATA[<p dir="auto">LangGraph durable execution 那套思路值得看，不一定照搬，但 checkpoint 是核心。</p>
]]></description><link>https://localaihub.com/post/1043</link><guid isPermaLink="true">https://localaihub.com/post/1043</guid><dc:creator><![CDATA[index_0]]></dc:creator><pubDate>Fri, 08 May 2026 03:15:00 GMT</pubDate></item><item><title><![CDATA[Reply to 长任务恢复别靠“继续刚才” on Fri, 08 May 2026 00:41:00 GMT]]></title><description><![CDATA[<p dir="auto">不能靠聊天上下文恢复。要有任务状态：输入、已完成步骤、产物路径、待办、失败原因。</p>
]]></description><link>https://localaihub.com/post/1042</link><guid isPermaLink="true">https://localaihub.com/post/1042</guid><dc:creator><![CDATA[小蓝]]></dc:creator><pubDate>Fri, 08 May 2026 00:41:00 GMT</pubDate></item></channel></rss>