
{"id":9219,"date":"2026-06-17T07:17:58","date_gmt":"2026-06-16T23:17:58","guid":{"rendered":"https:\/\/infernews.com\/blog\/visualclaw\/"},"modified":"2026-06-17T07:29:06","modified_gmt":"2026-06-16T23:29:06","slug":"visualclaw","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/visualclaw\/","title":{"rendered":"VisualClaw\uff1a\u6703\u9032\u5316\u7684\u8996\u89ba\u4ee3\u7406"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/visualclaw_icon-4.png\" alt=\"VisualClaw icon\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6642\u4e0d\u5c11 vision-language agents \u4e3b\u8981\u9760\u56fa\u5b9a\u63d0\u793a\u3001\u6574\u6bb5\u5f71\u7247\u9001\u5165\u6a21\u578b\uff0c\u6216\u8005\u4e00\u6b21\u6027\u5de5\u5177\u8abf\u7528\u53bb\u5b8c\u6210\u4efb\u52d9\uff1b\u4f5c\u8005\u8a8d\u70ba\u9019\u7a2e fixed scaffold \u5bb9\u6613\u4ee4\u5ef6\u9072\u3001\u6210\u672c\u540c\u63d0\u793a\u9577\u5ea6\u4e00\u9f4a\u4e0a\u5347\uff0c\u800c\u4e14\u90e8\u7f72\u5f8c\u5e7e\u4e4e\u5514\u6703\u81ea\u5df1\u7d2f\u7a4d\u7d93\u9a57\u3002VisualClaw \u56e0\u800c\u63d0\u51fa\u4e00\u500b modular multimodal agent system\uff0c\u593e\u5728 Claude Code\u3001Codex\u3001OpenClaw \u540c\u4e0a\u6e38 LLM providers \u4e4b\u9593\uff0c\u52a0\u5165 retrieved memory\u3001skill bank\u3001self-evolution\u3001video processing \u540c evaluation hooks\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u9805\u76ee\u5c6c\u65bc<strong>\u6846\u67b6\u517c\u5de5\u5177\u578b\u57fa\u5efa<\/strong>\uff0c\u91cd\u9ede\u5514\u4fc2\u518d\u8a13\u7df4\u4e00\u500b\u65b0\u6a21\u578b\uff0c\u800c\u4fc2\u5e6b Computer-use agents\u3001tool-using vision-language agents \u66f4\u6709\u6548\u8655\u7406\u5f71\u7247\u3001\u5de5\u4f5c\u7a7a\u9593\u540c\u591a\u8f2a\u4efb\u52d9\u3002\u914d\u5957\u7684 VisualClawArena \u5247\u662f\u7368\u7acb benchmark\/data release\uff0c\u6536\u9304 200 \u500b scenario\uff0c\u8981\u6c42\u4ee3\u7406\u771f\u6b63\u5728 workspace \u5167\u7528\u5f71\u7247\u8b49\u64da\u3001\u6587\u4ef6\u3001\u52d5\u614b\u66f4\u65b0\u8207 executable checks \u89e3\u984c\uff0c\u88dc\u56de\u4e00\u822c video-QA benchmark \u5c0d\u5de5\u5177\u4f7f\u7528\u60c5\u5883\u6e2c\u8a66\u4e0d\u8db3\u7684\u7f3a\u53e3\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u628a\u5b83\u7576\u6210 gateway proxy\uff0c\u7528\u719f\u6089\u7684 OpenAI \u6216 Anthropic \u76f8\u5bb9\u4ecb\u9762\u63a5\u5230\u73fe\u6709 agent workflow\uff0c\u518d\u9010\u6b65\u6253\u958b skills\u3001memory\u3001video cascade \u7b49\u6a21\u7d44\u3002\u5c0d\u5df2\u7d93\u5728\u7528 Claude Code \u6216 Codex \u7684\u5718\u968a\uff0c\u9019\u7a2e\u63a5\u6cd5\u5c24\u5176\u5be6\u7528\uff0c\u56e0\u70ba\u6bcb\u9808\u7531\u96f6\u91cd\u5beb\u6574\u500b\u4ee3\u7406\u6d41\u7a0b\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6838\u5fc3\u7d50\u69cb\u6e05\u695a<\/strong>\uff1agateway design \u52a0\u4e0a pre\/post hooks\uff0c\u65b9\u4fbf\u5728\u63d0\u793a\u524d\u5f8c\u63d2\u5165\u6280\u80fd\u3001\u8a18\u61b6\u8207\u8a55\u6e2c\u3002<\/li>\n\n\n\n<li><strong>\u5275\u65b0\u4f4d\u5177\u9ad4<\/strong>\uff1a\u4ee5 hybrid encoding \u8655\u7406 streaming video\uff0c\u914d\u5408 cascaded gate\u3001keyframe context\u3001hot\/cold top-k injection\uff0c\u76ee\u6a19\u662f\u58d3\u4f4e\u6210\u672c\u800c\u4fdd\u7559\u95dc\u9375\u8996\u89ba\u8b49\u64da\u3002<\/li>\n\n\n\n<li><strong>\u6703\u7d2f\u7a4d\u7d93\u9a57<\/strong>\uff1aself-evolution \u6703\u6839\u64da\u5931\u6557\u56de\u5408\u8207\u6210\u529f\u8a18\u61b6\u66f4\u65b0 skill bank\uff0c\u5514\u4fc2\u6bcf\u6b21\u90fd\u7531\u540c\u4e00\u5957\u975c\u614b\u63d0\u793a\u958b\u59cb\u3002<\/li>\n\n\n\n<li><strong>\u8a55\u6e2c\u5514\u53ea\u770b\u7b54\u5c0d<\/strong>\uff1aVisualClawArena \u628a\u591a\u8f2a\u554f\u984c\u3001\u5f71\u7247\u3001workspace \u64cd\u4f5c\u540c\u8a55\u4f30\u8f38\u51fa\u7d81\u57cb\u4e00\u9f4a\uff0c\u66f4\u63a5\u8fd1\u771f\u5be6\u4ee3\u7406\u4efb\u52d9\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728 4 \u500b video-QA benchmarks\u30012 \u500b VLMs\uff08Gemini 3 Flash\u3001GPT-5.2\uff09\u4e0a\uff0cVisualClaw \u5c0d full-frame upload \u7684\u6bcf\u984c API \u6210\u672c\u5e73\u5747\u53ef\u6e1b 98%\uff0c\u76f8\u5c0d offline uniform 8 frame baseline \u4ea6\u518d\u964d 25.9%\uff0c\u540c\u6642\u591a\u6578\u8a2d\u5b9a\u4e0b\u6e96\u78ba\u7387\u6709\u63d0\u5347\u3002\u5230 VisualClawArena\uff0c\u914d\u5408 Codex\uff08GPT-5.5\uff09\u8207 Claude Code\uff08Sonnet 4.6\uff09\u7b49 backend\uff0cmacro accuracy \u5206\u5225\u63d0\u5347 2.9% \u540c 3.2%\uff1b\u5982\u679c\u4f60\u95dc\u5fc3\u7684\u662f\u53ef\u63d2\u62d4\u4ee3\u7406\u57fa\u5efa\u3001\u591a\u6a21\u614b\u5de5\u4f5c\u6d41\uff0c\u6216\u8005\u60f3\u70ba live vision \u8207\u5f71\u7247\u4efb\u52d9\u52a0\u4e0a\u53ef\u6301\u7e8c\u9032\u5316\u80fd\u529b\uff0c\u9019\u500b\u9805\u76ee\u6bd4\u55ae\u7d14\u6a21\u578b\u5305\u88dd\u5668\u66f4\u6709\u7814\u7a76\u540c\u843d\u5730\u50f9\u503c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/UCSC-VLAA\/VisualClaw\" rel=\"noopener noreferrer\">https:\/\/github.com\/UCSC-VLAA\/VisualClaw<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9805\u76ee\uff1a<\/strong> <a href=\"https:\/\/ucsc-vlaa.github.io\/VisualClaw\/\" rel=\"noopener noreferrer\">https:\/\/ucsc-vlaa.github.io\/VisualClaw\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>VisualClaw \u4e0d\u53ea\u662f\u63a5\u99c1\u6a21\u578b\u7684\u5de5\u5177\uff0c\u9084\u628a\u8a18\u61b6\u3001\u6280\u80fd\u8207\u5f71\u7247\u8655\u7406\u4e32\u6210\u53ef\u6301\u7e8c\u6539\u826f\u7684\u4ee3\u7406\u7cfb\u7d71\u3002<\/p>\n","protected":false},"author":8,"featured_media":9218,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,140,153,116,137,31,38,162,158,166,159,119,76,127,146,191,190,201,197,196],"tags":[],"class_list":["post-9219","post","type-post","status-publish","format-standard","hentry","category-133","category-gemini","category-openai","category-agentic","category-api","category-video","category-38","category-ai-productions","category-ide","category-mac","category-vibe-coding","category-119","category-76","category-127","category-146","category-anthropic","category-openclaw","category-skill-","category-framework","category-196"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9219","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=9219"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9219\/revisions"}],"predecessor-version":[{"id":9220,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9219\/revisions\/9220"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9218"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9219"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9219"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9219"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}