{"id":11448,"date":"2026-08-26T13:54:43","date_gmt":"2026-08-26T05:54:43","guid":{"rendered":"https:\/\/infernews.com\/blog\/autosaddler\/"},"modified":"2026-08-26T13:58:51","modified_gmt":"2026-08-26T05:58:51","slug":"autosaddler","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/autosaddler\/","title":{"rendered":"AutoSaddler \u91cd\u65b0\u6574\u7406 LLM Agent Harness"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-a3a42edd38af.jpg\" alt=\"AutoSaddler Overall Framework\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Microsoft \u7814\u7a76\u5718\u968a\u806f\u540c POSTECH\u3001KAIST \u53ca\u5357\u65b9\u79d1\u6280\u5927\u5b78\u958b\u767c AutoSaddler\uff0c\u91dd\u5c0d\u9577\u6d41\u7a0b AI Agent \u5bb9\u6613\u56e0\u5c0f\u932f\u8aa4\u7d2f\u7a4d\u800c\u5931\u6557\u7684\u554f\u984c\uff0c\u5efa\u7acb\u81ea\u52d5\u512a\u5316 harness \u7684\u65b9\u6cd5\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">AutoSaddler \u4fc2\u4e00\u500b\u7528\u569f\u81ea\u52d5\u512a\u5316 LLM agent harness \u5605\u958b\u6e90\u9805\u76ee\uff0c\u8655\u7406\u5605\u4fc2\u9577\u6d41\u7a0b\u4efb\u52d9\u5165\u9762\uff0cagent \u56e0\u70ba\u63d0\u793a\u8a5e\u3001\u5de5\u5177\u6216\u63a7\u5236\u908f\u8f2f\u6709\u5c11\u5c11\u504f\u5dee\uff0c\u5c31\u4e00\u8def\u7d2f\u7a4d\u932f\u8aa4\u800c\u5931\u6557\u5462\u500b\u554f\u984c\u3002\u4f62\u5514\u4fc2\u55ae\u9760\u53cd\u8986\u6539 prompt\uff0c\u800c\u4fc2\u5c07 harness \u7576\u6210\u7a0b\u5f0f\u53bb\u8a3a\u65b7\u540c\u4fee\u88dc\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f62\u6703\u5206\u6790\u57f7\u884c trace\uff0c\u627e\u51fa\u5931\u6557\u6839\u6e90\uff0c\u518d\u6309\u65e2\u5b9a\u88dc\u4e01\u5206\u985e\u53bb\u6539 prompts\u3001tools\u3001middleware \u540c agent-loop logic\u3002\u8cc7\u6599\u593e\u540c \u4f62\u7528 durable\u3001plugin-based \u65b9\u5f0f\u8a18\u9304\u72c0\u614b\uff0c\u9069\u5408\u8981\u53cd\u8986\u8a66\u9a57\u3001\u56de\u5fa9\u540c\u8ffd\u8e64\u4fee\u6539\u6b77\u7a0b\u5605\u5de5\u4f5c\u6d41\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u540c\u985e\u65b9\u6cd5\u901a\u5e38\u53ea\u6539\u5c11\u91cf\u63d0\u793a\u8a5e\uff0cAutoSaddler \u5c31\u5c07\u641c\u7d22\u7bc4\u570d\u64f4\u5927\u5230\u6574\u500b harness\uff0c\u4ef2\u6703\u7528 validation \u53bb\u6311\u9078\u8f03\u80fd\u6cdb\u5316\u5605\u66f4\u65b0\u3002\u521d\u6b65\u7d50\u679c\u986f\u793a\uff0c\u4f62\u55ba GAIA2\u3001SWE-Bench Pro \u540c Terminal-Bench 2.0 \u90fd\u5e36\u4f86\u7d04 9 \u5230 10 \u500b\u767e\u5206\u9ede\u5605 Pass@1 \u63d0\u5347\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5c0d\u505a agent \u7cfb\u7d71\u3001\u8a55\u6e2c\u6846\u67b6\u6216\u8005\u81ea\u52d5\u5316\u5de5\u4f5c\u6d41\u5605\u5718\u968a\u7279\u5225\u6709\u7528\uff0c\u56e0\u70ba\u4f62\u8655\u7406\u5514\u4fc2\u55ae\u4e00\u6a21\u578b\u8f38\u51fa\uff0c\u800c\u4fc2\u6574\u5957\u57f7\u884c\u74b0\u5883\u9ede\u6a23\u66f4\u7a69\u5b9a\u3002\u9805\u76ee\u8981\u6c42 Python 3.12-3.14\u3001uv \u540c Git\uff0c\u5b98\u65b9\u5efa\u8b70\u7528 <code>uv run<\/code> \u53bb\u57f7\u884c Python \u6307\u4ee4\uff0c\u4ee3\u8868\u4f62\u504f\u5411\u53ef\u91cd\u73fe\u540c\u53ef\u8ffd\u8e64\u5605\u7814\u7a76\u8207\u5de5\u7a0b\u6574\u5408\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5 failure trace \u8a3a\u65b7\u554f\u984c\uff0c\u5514\u4fc2\u6de8\u4fc2\u505a\u8868\u9762\u53cd\u601d<\/li>\n\n\n\n<li>\u6539\u52d5\u7bc4\u570d\u5305\u62ec prompt\u3001\u5de5\u5177\u3001middleware \u540c agent loop<\/li>\n\n\n\n<li>\u6703\u9a57\u8b49\u66f4\u65b0\u55ba\u5176\u4ed6\u60c5\u5883\u53ef\u5514\u53ef\u4ee5\u4fdd\u6301\u6548\u679c<\/li>\n\n\n\n<li>\u521d\u6b65 benchmark \u7d50\u679c\u986f\u793a\u6709\u660e\u986f\u63d0\u5347<\/li>\n\n\n\n<li>\u9069\u5408\u8981\u6301\u7e8c\u8abf\u6574 LLM agent harness \u5605\u5718\u968a<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/autosaddler-projectpage.github.io\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/microsoft\/AutoSaddler\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>AutoSaddler \u6703\u5f9e\u57f7\u884c\u5931\u6557\u7d00\u9304\u5165\u624b\uff0c\u5e6b LLM agent \u81ea\u52d5\u8abf\u6574\u63d0\u793a\u8a5e\u3001\u5de5\u5177\u540c\u4e2d\u4ecb\u5c64\u3002\u5b83\u5514\u53ea\u4fee\u4e00\u689d\u8def\u5f91\uff0c\u4ef2\u6703\u6aa2\u67e5\u6539\u52d5\u80fd\u5426\u63a8\u5ee3\u5230\u5176\u4ed6\u60c5\u5883\u3002<\/p>\n","protected":false},"author":8,"featured_media":11447,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,116,173,197,38,152,199],"tags":[],"class_list":["post-11448","post","type-post","status-publish","format-standard","hentry","category-133","category-agentic","category-173","category-framework","category-38","category-python","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11448","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=11448"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11448\/revisions"}],"predecessor-version":[{"id":11449,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11448\/revisions\/11449"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11447"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11448"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11448"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11448"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}