<?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[Claude 和 GPT 做长任务，怎么避免中途跑偏？]]></title><description><![CDATA[<p dir="auto">我让 Claude/GPT 做一整天的资料整理和代码修改，前面很好，后面会忘目标或者开始做无关优化。大家怎么控？</p>
]]></description><link>https://localaihub.com/topic/89/claude-和-gpt-做长任务-怎么避免中途跑偏</link><generator>RSS for Node</generator><lastBuildDate>Wed, 03 Jun 2026 18:50:48 GMT</lastBuildDate><atom:link href="https://localaihub.com/topic/89.rss" rel="self" type="application/rss+xml"/><pubDate>Tue, 05 May 2026 10:59:00 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 12:39:00 GMT]]></title><description><![CDATA[<p dir="auto">我准备加任务状态文件、阶段验收、可写路径限制。长上下文只当辅助。</p>
]]></description><link>https://localaihub.com/post/618</link><guid isPermaLink="true">https://localaihub.com/post/618</guid><dc:creator><![CDATA[小唐]]></dc:creator><pubDate>Wed, 06 May 2026 12:39:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 09:34:00 GMT]]></title><description><![CDATA[<p dir="auto">这就是范围边界没锁。工具层面也能限制可写路径。</p>
]]></description><link>https://localaihub.com/post/617</link><guid isPermaLink="true">https://localaihub.com/post/617</guid><dc:creator><![CDATA[小陈在改bug]]></dc:creator><pubDate>Wed, 06 May 2026 09:34:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 08:12:00 GMT]]></title><description><![CDATA[<p dir="auto">我之前让它跑着跑着改了 README 和样式，原任务是修 API bug。</p>
]]></description><link>https://localaihub.com/post/616</link><guid isPermaLink="true">https://localaihub.com/post/616</guid><dc:creator><![CDATA[小唐]]></dc:creator><pubDate>Wed, 06 May 2026 08:12:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 07:22:00 GMT]]></title><description><![CDATA[<p dir="auto">还有人工 checkpoint。生产级代码 agent 不应该无限自治到推送。</p>
]]></description><link>https://localaihub.com/post/615</link><guid isPermaLink="true">https://localaihub.com/post/615</guid><dc:creator><![CDATA[陈一]]></dc:creator><pubDate>Wed, 06 May 2026 07:22:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 04:39:00 GMT]]></title><description><![CDATA[<p dir="auto">我们把“不得做无关重构”写在验收里，比写在普通提示里有效。</p>
]]></description><link>https://localaihub.com/post/614</link><guid isPermaLink="true">https://localaihub.com/post/614</guid><dc:creator><![CDATA[别问问就是缓存]]></dc:creator><pubDate>Wed, 06 May 2026 04:39:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 03:01:00 GMT]]></title><description><![CDATA[<p dir="auto">看窗口和任务阶段。每完成一个可验证阶段就压缩一次，比按轮数压缩更自然。</p>
]]></description><link>https://localaihub.com/post/613</link><guid isPermaLink="true">https://localaihub.com/post/613</guid><dc:creator><![CDATA[nora]]></dc:creator><pubDate>Wed, 06 May 2026 03:01:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Wed, 06 May 2026 00:26:00 GMT]]></title><description><![CDATA[<p dir="auto">上下文压缩频率怎么定？</p>
]]></description><link>https://localaihub.com/post/612</link><guid isPermaLink="true">https://localaihub.com/post/612</guid><dc:creator><![CDATA[阿树]]></dc:creator><pubDate>Wed, 06 May 2026 00:26:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 23:54:00 GMT]]></title><description><![CDATA[<p dir="auto">Claude 的长上下文读材料强，但长任务执行仍然会漂。GPT 也一样。长上下文解决输入，不解决目标管理。</p>
]]></description><link>https://localaihub.com/post/611</link><guid isPermaLink="true">https://localaihub.com/post/611</guid><dc:creator><![CDATA[leaf_1997]]></dc:creator><pubDate>Tue, 05 May 2026 23:54:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 22:40:00 GMT]]></title><description><![CDATA[<p dir="auto">可以，但 todo 要和工具结果绑定。比如“测试通过”必须来自命令输出，不是模型自评。</p>
]]></description><link>https://localaihub.com/post/610</link><guid isPermaLink="true">https://localaihub.com/post/610</guid><dc:creator><![CDATA[zeroOne]]></dc:creator><pubDate>Tue, 05 May 2026 22:40:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 20:16:00 GMT]]></title><description><![CDATA[<p dir="auto">模型自己写 todo 可以吗？</p>
]]></description><link>https://localaihub.com/post/609</link><guid isPermaLink="true">https://localaihub.com/post/609</guid><dc:creator><![CDATA[小满]]></dc:creator><pubDate>Tue, 05 May 2026 20:16:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 18:37:00 GMT]]></title><description><![CDATA[<p dir="auto">我们做 agent 时会保存 task brief：用户原始目标、禁止事项、验收标准、当前 repo 状态。每次新循环都注入。</p>
]]></description><link>https://localaihub.com/post/608</link><guid isPermaLink="true">https://localaihub.com/post/608</guid><dc:creator><![CDATA[melo]]></dc:creator><pubDate>Tue, 05 May 2026 18:37:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 17:04:00 GMT]]></title><description><![CDATA[<p dir="auto">每个阶段都要重新给它目标、约束、当前事实。上下文压缩不是把所有历史压成散文。</p>
]]></description><link>https://localaihub.com/post/607</link><guid isPermaLink="true">https://localaihub.com/post/607</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Tue, 05 May 2026 17:04:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 16:26:00 GMT]]></title><description><![CDATA[<p dir="auto">代码任务尤其要限制范围。模型看到顺手能改的东西，会忍不住“清理一下”。</p>
]]></description><link>https://localaihub.com/post/606</link><guid isPermaLink="true">https://localaihub.com/post/606</guid><dc:creator><![CDATA[小陈在改bug]]></dc:creator><pubDate>Tue, 05 May 2026 16:26:00 GMT</pubDate></item><item><title><![CDATA[Reply to Claude 和 GPT 做长任务，怎么避免中途跑偏？ on Tue, 05 May 2026 14:06:00 GMT]]></title><description><![CDATA[<p dir="auto">长任务要有外部任务状态，别只靠对话记忆。计划、已完成、待确认、风险，都写到可读状态里。</p>
]]></description><link>https://localaihub.com/post/605</link><guid isPermaLink="true">https://localaihub.com/post/605</guid><dc:creator><![CDATA[JasonL]]></dc:creator><pubDate>Tue, 05 May 2026 14:06:00 GMT</pubDate></item></channel></rss>