{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-08-07T02:00:35.291Z","headline":"Microsoft 的 SkillOpt 证明优化后的智能体技能工件可在不同模型规模及 Codex 与 Claude Code 之间迁移","description":"Microsoft 与上海交大、同济、复旦团队提出的 SkillOpt 通过文本空间优化训练单一技能文档，冻结目标模型，使优化后的技能工件可跨模型规模和跨工具链迁移。在 Codex 上优化的 SpreadsheetBench 技能部署到 Claude Code 后得分 81.8，超过后者自行训练技能得到的 80.4。全部 4 项跨模型、4 项跨工具链和 3 项跨基准迁移结果均高于目标的无技能基线。","url":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/","mainEntityOfPage":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/","datePublished":"2026-08-06T00:37:42.000Z","dateModified":"2026-08-06T00:37:42.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability","https://aihot.virxact.com/items/cmsgsgz530fluro5qu0vnw8s0"],"canonicalUrl":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/","directAnswer":{"@type":"Answer","text":"SkillOpt 以冻结目标模型、优化单一自然语言技能文档的方式提升任务表现。材料显示，其技能可在同一 GPT 系列的不同规模间迁移，也可从 Codex 迁移至 Claude Code，且所列迁移结果均未低于目标的无技能基线。","url":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/","dateCreated":"2026-08-06T00:37:42.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"marktechpost.com source article","url":"https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability","datePublished":"2026-08-06T00:37:42.000Z","provider":{"@type":"Organization","name":"marktechpost.com","url":"https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmsgsgz530fluro5qu0vnw8s0","datePublished":"2026-08-06T00:37:42.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmsgsgz530fluro5qu0vnw8s0"}}],"aggregationSource":"MarkTechPost（RSS）","originalPublisher":{"name":"marktechpost.com","url":"https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability"},"geoDeepAnswer":null,"article":{"id":"cmsgsgz530fluro5qu0vnw8s0","slug":"cmsgsgz530fluro5qu0vnw8s0","url":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/","title":"Microsoft 的 SkillOpt 证明优化后的智能体技能工件可在不同模型规模及 Codex 与 Claude Code 之间迁移","title_en":"Microsoft's SkillOpt Shows Optimized Agent Skill Artifacts Transfer Across Model Scales and Between Codex and Claude Code Harnesses","summary":"Microsoft 与上海交大、同济、复旦团队提出的 SkillOpt 通过文本空间优化训练单一技能文档，冻结目标模型，使优化后的技能工件可跨模型规模和跨工具链迁移。在 Codex 上优化的 SpreadsheetBench 技能部署到 Claude Code 后得分 81.8，超过后者自行训练技能得到的 80.4。全部 4 项跨模型、4 项跨工具链和 3 项跨基准迁移结果均高于目标的无技能基线。","source":"MarkTechPost（RSS）","sourceUrl":"https://www.marktechpost.com/2026/08/05/microsoft-skillopt-agent-skill-transfer-portability","aiHotUrl":"https://aihot.virxact.com/items/cmsgsgz530fluro5qu0vnw8s0","publishedAt":"2026-08-06T00:37:42.000Z","category":"论文研究","score":70,"selected":true,"articleBody":["SkillOpt：https://github.com/microsoft/SkillOpt is a text-space optimizer developed by a team of researchers from Microsoft, Shanghai Jiao Tong University, Tongji University, and Fudan University.","SkillOpt trains a single natural-language skill document while the target model stays frozen. An optimizer model reads scored rollouts and proposes bounded add/delete/replace edits. A held-out selection split accepts an edit only when the score strictly improves. The exported artifact is one file, best_skill.md .","The transfer tables report three columns. Baseline is the target’s no-skill score. Direct is SkillOpt trained in-domain on that exact target. Transferred applies a skill trained elsewhere, with no further optimization.","The useful comparison is not transferred versus direct. It is how much of the in-domain gain survives the move.","Skills were trained on GPT-5.4 and deployed on smaller variants.","Two rows deserve attention. SpreadsheetBench on GPT-5.4-mini keeps 82% of the in-domain gain. That is close to free reuse. The LiveMath row on GPT-5.4-nano is stranger: the transferred skill scores 28.8 against an in-domain SkillOpt result of 27.2. The paper reads this as evidence that some learned procedures are target-model agnostic.","The GPT-5.4-nano SpreadsheetBench row is the weak one at 16%. Retention is not uniform, and the paper does not claim it is. Its stated bound is narrower: no row falls below the target’s no-skill baseline.","Note the scope. All four rows stay inside one GPT family. Cross-family transfer, such as GPT to Qwen：https://qwen.ai/blog?id=qwen3.5, is not tested.","This is the section that matters most for deployment. All rows use GPT-5.5.","The first row is the headline. A skill optimized inside Codex：https://openai.com/index/introducing-codex/ lifted Claude Code：https://www.anthropic.com/claude-code from 22.1 to 81.8. That slightly exceeds the 80.4 Claude Code reached by training its own skill from scratch.","The two harnesses expose different tool and file APIs and different command surfaces. A skill that survives that shift is not encoding command recipes. The research paper：https://arxiv.org/pdf/2605.23904 attributes SpreadsheetBench’s portability to workbook-level procedures: structure-first inspection, formula-aware verification, and static-value materialization. Those hold regardless of which CLI runs the Python.","LiveMath tells the opposite story. Codex → Claude Code retains only 10% of the in-domain gain. The asymmetry is worth sitting with. Procedural skills — how to inspect, verify, and format — appear to be the portable class. Reasoning-heavy skills appear more tied to their training environment.","There is no Direct column here. No in-domain SkillOpt run on Omni-MATH is reported, so the comparison is against no-skill only. Gains are positive across all three model scales but small. The research paper’s reading is that the skill retained reusable mathematical procedure after both the test instances and the answer-format conventions changed.","The mechanism is stated plainly in the research paper. All three execution modes: direct chat, Codex, Claude Code – consume the same best_skill.md file format. That shared contract is what makes the cross-harness experiment possible in the first place.","The Codex harness renders the current skill to a per-task SKILL.md alongside task files, then reads back a compact execution trace. The Claude Code harness mirrors the same workspace contract through the claude CLI. Neither harness gets a bespoke skill format.","The artifact’s shape supports portability too. Final skills run 379 to 1,995 tokens across the six benchmarks, with a median near 920. They are assembled from 1 to 4 accepted edits. The paper’s Figure 4 samples one learned rule per benchmark, and all are procedural rather than instance-specific. The SpreadsheetBench rule, verbatim: inspect workbook structure and formulas, then write evaluated static values across the full requested target range instead of relying on Excel recalculation.","Training cost is paid once, offline, and measured. The research paper reports 0.6M to 46.4M training tokens per absolute test point, depending on benchmark. SpreadsheetBench sits at 0.6M per point; DocVQA at 46.4M. The optimizer model runs only during training and adds zero inference-time calls at deployment.","If a skill trained in one harness holds up in another, that one-time cost spreads across environments. The Codex → Claude Code SpreadsheetBench result is the existence proof. It also implies you can optimize where tooling is cheapest and deploy where the product lives.","The audit angle is separate and underrated. The deployed artifact is a text file a domain practitioner can read in minutes. Every change to it is traceable: each step records an edit_apply_report.json with per-edit accept and skip status. Portability plus inspectability is a different operational posture than shipping fine-tuned weights.","Resources: Paper：https://arxiv.org/abs/2605.23904, GitHub：https://github.com/microsoft/SkillOpt, Project page：https://microsoft.github.io/SkillOpt/, Docs：https://github.com/microsoft/SkillOpt/blob/main/docs/index.md, PyPI：https://pypi.org/project/skillopt/ and Demo video：https://youtu.be/JUBMDTCiM0M","Baselines referenced: GEPA：https://arxiv.org/abs/2507.19457, TextGrad：https://arxiv.org/abs/2406.07496, EvoSkill：https://arxiv.org/abs/2603.02766 and Trace2Skill：https://arxiv.org/abs/2603.25158","Benchmarks referenced: SearchQA：https://arxiv.org/abs/1704.05179, SpreadsheetBench：https://arxiv.org/abs/2406.14991, DocVQA：https://arxiv.org/abs/2007.00398, LiveMathematicianBench：https://arxiv.org/abs/2604.01754 and ALFWorld：https://openreview.net/forum?id=0IOX0YcCdTn","Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.","[FREE GUIDE] Securing AI Agents, MCP Servers & LLM Apps ：https://pxllnk.co/lxn88m"],"articleImages":[{"sourceUrl":"https://www.marktechpost.com/wp-content/uploads/2019/06/Screen-Shot-2021-09-14-at-9.02.24-AM-300x300.png","alt":"","afterParagraph":21,"url":"/media/articles/cmsgsgz530fluro5qu0vnw8s0/787a6d54564e8e19.webp"},{"sourceUrl":"https://www.marktechpost.com/wp-content/uploads/2026/08/blog6176-14-100x70.png","alt":"End-to-End Bayesian Marketing Mix Modeling with Google Meridian: Media Measurement, ROI Analysis, and Budget Optimization","afterParagraph":22,"url":"/media/articles/cmsgsgz530fluro5qu0vnw8s0/f8a290c04ce05d94.png"},{"sourceUrl":"https://www.marktechpost.com/wp-content/uploads/2026/08/blog6176-13-100x70.png","alt":"Meta AI Releases Muse Code","afterParagraph":22,"url":"/media/articles/cmsgsgz530fluro5qu0vnw8s0/a92fcb4264c67d98.png"},{"sourceUrl":"https://www.marktechpost.com/wp-content/uploads/2026/08/blog6176-12-100x70.png","alt":"NVIDIA Releases Alpamayo 2 Super","afterParagraph":22,"url":"/media/articles/cmsgsgz530fluro5qu0vnw8s0/840d7d20551c1f7b.webp"},{"sourceUrl":"https://www.marktechpost.com/wp-content/uploads/2026/08/blog6176-1-1-100x70.png","alt":"Pixel-Native RAG: A Practical Guide to Visual Document Indexing","afterParagraph":22,"url":"/media/articles/cmsgsgz530fluro5qu0vnw8s0/5c8b01b8b70a10f0.webp"}],"mediaStatus":"ok","articleBodyZh":["SkillOpt：https://github.com/microsoft/SkillOpt 是一个文本空间优化器，由来自微软、上海交通大学、同济大学和复旦大学的研究团队开发。","SkillOpt 在目标模型保持冻结的情况下训练单个自然语言技能文档。优化器模型读取评分回合并提出有限的添加/删除/替换编辑。只有当评分严格提高时，保留的选择划分才会接受编辑。导出的产物是一个文件，best_skill.md。","迁移表格报告三列。Baseline 是目标模型的无技能分数。Direct 是在该目标域内训练的 SkillOpt。Transferred 应用在其他地方训练的技能，没有进一步优化。","有用的比较不是 transferred 与 direct 的对比，而是域内增益在迁移后能保留多少。","技能在 GPT-5.4 上训练，并部署到较小的变体上。","两行值得关注。GPT-5.4-mini 上的 SpreadsheetBench 保留了 82% 的域内增益。这接近于免费复用。GPT-5.4-nano 上的 LiveMath 行更奇怪：迁移的技能得分为 28.8，而域内 SkillOpt 结果为 27.2。论文将其解读为某些学习的程序与目标模型无关的证据。","GPT-5.4-nano 上的 SpreadsheetBench 行较弱，仅为 16%。保留效果并不均匀，论文也没有声称均匀。其声明的界限更窄：没有任何一行低于目标模型的无技能基线。","注意范围。所有四行都在同一 GPT 系列内。跨系列迁移，例如 GPT 到 Qwen：https://qwen.ai/blog?id=qwen3.5，尚未测试。","这是部署中最重要的部分。所有行均使用 GPT-5.5。","第一行是重点。在 Codex：https://openai.com/index/introducing-codex/ 内优化的技能将 Claude Code：https://www.anthropic.com/claude-code 从 22.1 提升到 81.8。略高于 Claude Code 通过从头训练自身技能达到的 80.4。","这两种工具架构暴露出不同的工具和文件 API 以及不同的命令界面。能够在这种变化下仍然生存的技能，并不是编码命令配方。研究论文：https://arxiv.org/pdf/2605.23904 将 SpreadsheetBench 的可移植性归因于工作簿级的过程：先结构化检查、识别公式的验证、以及静态值物化。这些内容不受运行 Python 的 CLI 类型影响。","LiveMath 展示了相反的情况。Codex → Claude Code 只保留了领域内提升的 10%。这种不对称值得关注。流程型技能——如何检查、验证和格式化——似乎是可移植的类别。依赖推理的技能则更依赖于其训练环境。","这里没有“直接”列。没有报告 Omni-MATH 上的领域内 SkillOpt 运行，因此比较对象仅为无技能情况。三种模型规模的增益都是正的，但很小。研究论文得出的结论是，技能在测试实例和答案格式规范变化之后，仍能保留可复用的数学过程。","研究论文中明确说明了机制。三种执行模式：直接聊天、Codex、Claude Code——都使用相同的 best_skill.md 文件格式。正是这种共享契约，使跨工具架构实验成为可能。","Codex 架构将当前技能渲染到每个任务的 SKILL.md 文件中，与任务文件一起，然后读取紧凑的执行痕迹。Claude Code 架构通过 claude CLI 反映相同的工作空间契约。两种工具架构都没有使用定制的技能格式。","该成果的形式也支持可移植性。最终技能在六个基准测试中运行时长度在 379 到 1,995 个 token 之间，中位数约 920。它们由 1 到 4 次通过审核的编辑组合而成。论文的图 4 从每个基准测试中抽取了一条学习规则，且全部为流程型，而非特定实例的规则。SpreadsheetBench 规则逐字原文：检查工作簿结构和公式，然后在整个请求的目标范围内写入评估后的静态值，而不是依赖 Excel 的重新计算。","训练成本一次性支付，离线计算，并且可测量。研究论文报告每个绝对测试点的训练标记数为0.6M到46.4M，取决于基准。SpreadsheetBench为每点0.6M；DocVQA为46.4M。优化器模型仅在训练期间运行，在部署时不会增加推理调用。","如果在一个环境中训练的技能在另一个环境中也能发挥作用，那么一次性成本就可以分摊到多个环境中。Codex → Claude Code SpreadsheetBench的结果就是存在性证明。这也意味着你可以在工具成本最低的地方进行优化，并在产品所在的地方部署。","审计角度是独立且被低估的。部署的产物是一个文本文件，领域从业者可以在几分钟内阅读。对其每一次的修改都是可追踪的：每一步都会记录一个 edit_apply_report.json，其中包含每次编辑的接受和跳过状态。可迁移性加上可检查性，是与仅发布微调权重不同的操作态度。","资源：论文：https://arxiv.org/abs/2605.23904，GitHub：https://github.com/microsoft/SkillOpt，项目页面：https://microsoft.github.io/SkillOpt/，文档：https://github.com/microsoft/SkillOpt/blob/main/docs/index.md，PyPI：https://pypi.org/project/skillopt/，演示视频：https://youtu.be/JUBMDTCiM0M","参考的基线：GEPA：https://arxiv.org/abs/2507.19457，TextGrad：https://arxiv.org/abs/2406.07496，EvoSkill：https://arxiv.org/abs/2603.02766 以及 Trace2Skill：https://arxiv.org/abs/2603.25158","参考的基准：SearchQA：https://arxiv.org/abs/1704.05179, SpreadsheetBench：https://arxiv.org/abs/2406.14991, DocVQA：https://arxiv.org/abs/2007.00398, LiveMathematicianBench：https://arxiv.org/abs/2604.01754 和 ALFWorld：https://openreview.net/forum?id=0IOX0YcCdTn","Asif Razzaq是Marktechpost Media Inc.的首席执行官。作为一位富有远见的企业家和工程师，Asif致力于利用人工智能的潜力造福社会。他最近的工作是推出人工智能媒体平台Marktechpost，该平台以其对机器学习和深度学习新闻的深入覆盖而脱颖而出，内容既技术可靠又易于广大受众理解。该平台每月访问量超过200万，显示了其在受众中的受欢迎程度。","[免费指南] 保护 AI 代理、MCP 服务器和 LLM 应用程序：https://pxllnk.co/lxn88m"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"SkillOpt 以冻结目标模型、优化单一自然语言技能文档的方式提升任务表现。材料显示，其技能可在同一 GPT 系列的不同规模间迁移，也可从 Codex 迁移至 Claude Code，且所列迁移结果均未低于目标的无技能基线。","background":"该方法由 Microsoft、上海交大、同济和复旦研究团队提出。优化器读取带评分的运行结果，对技能文档执行受限的增删改；只有在留出选择集上得分严格提升时才接受编辑，最终导出单一的 best_skill.md 文件。","viewpoint":"Aioga 判断，这项研究的重点不是迁移成绩能否超过目标端直接优化，而是原有增益在迁移后保留多少。部分结果显示技能可能包含与特定目标模型无关的程序性经验，但不同任务和模型规模间的保留率并不一致。","implications":"对智能体开发而言，可迁移的文本技能工件可能降低针对每个模型或工具链重复优化的需求。值得关注的是，SpreadsheetBench 技能从 Codex 部署到 Claude Code 后得分为 81.8，高于后者自行训练技能的 80.4。","nextStep":"后续应重点验证跨模型家族迁移，因为现有跨规模测试仍局限于同一 GPT 家族，尚未测试 GPT 到 Qwen。也应继续比较不同任务的增益保留率，解释 GPT-5.4-nano 上 SpreadsheetBench 仅保留 16% 的情况。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-08-06T02:07:05.356Z","sourceHash":"e5222008d94a4cbc","review":{"approved":true,"groundedness":97,"clarity":94,"duplicationRisk":18,"blockingIssues":[],"notes":["“可能降低针对每个模型或工具链重复优化的需求”属于合理的前景判断，候选内容已使用“可能”作限定，未将其表述为既成事实。","可选措辞建议：将“后者自行训练技能”改为“在 Claude Code 目标端直接优化得到的技能”，可更准确对应来源中的 in-domain/direct 设定。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["论文研究","MarkTechPost（RSS）"],"translations":{"zh-CN":{"title":"Microsoft 的 SkillOpt 证明优化后的智能体技能工件可在不同模型规模及 Codex 与 Claude Code 之间迁移","summary":"Microsoft 与上海交大、同济、复旦团队提出的 SkillOpt 通过文本空间优化训练单一技能文档，冻结目标模型，使优化后的技能工件可跨模型规模和跨工具链迁移。在 Codex 上优化的 SpreadsheetBench 技能部署到 Claude Code 后得分 81.8，超过后者自行训练技能得到的 80.4。全部 4 项跨模型、4 项跨工具链和 3 项跨基准迁移结果均高于目标的无技能基线。","category":"论文研究","source":"marktechpost.com","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsoft 的 SkillOpt 证明优化后的智能体技能工件可在不同模型规模及 Codex 与 Claude Code 之间迁移 - Aioga AI资讯","description":"Microsoft 与上海交大、同济、复旦团队提出的 SkillOpt 通过文本空间优化训练单一技能文档，冻结目标模型，使优化后的技能工件可跨模型规模和跨工具链迁移。在 Codex 上优化的 SpreadsheetBench 技能部署到 Claude Code 后得分 81.8，超过后者自行训练技能得到的 80.4。全部 4 项跨模型、4 项跨工具链和 3...","url":"https://www.aioga.com/news/cmsgsgz530fluro5qu0vnw8s0/"},"en":{"title":"Microsoft's SkillOpt proves that optimized agent skill artifacts can migrate between different model sizes and Codex and Claude Code","summary":"Microsoft, together with teams from Shanghai Jiao Tong University, Tongji, and Fudan, proposed SkillOpt, which trains single-skill documents through text space optimization, freezes target models, and enables optimized skill artifacts to migrate across model scales and toolchains. The SpreadsheetBench skill optimized on Codex scored 81.8 when deployed to Claude Code, surpassing the 80.4 achieved by the latter's self-trained skills. All four cross-model, four cross-toolchain, and three cross-benchmark migration results were above the target's no-skills baseline.","category":"Research","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsoft's SkillOpt proves that optimized agent skill artifacts can migrate between different model sizes and Codex and Claude Code - Aioga AI News","description":"Microsoft, together with teams from Shanghai Jiao Tong University, Tongji, and Fudan, proposed SkillOpt, which trains single-skill documents through text space optimization, freeze...","url":"https://www.aioga.com/en/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:00.739Z"},"ja":{"title":"MicrosoftのSkillOptは、最適化されたエージェントスキルのアーティファクトが異なるモデルサイズやCodex、Claudeコード間で移行できることを証明しています","summary":"Microsoftは上海交通大学、通済大学、復旦のチームと共に、単一スキル文書をテキスト空間最適化で訓練し、ターゲットモデルを凍結し、最適化されたスキルアーティファクトをモデルスケールやツールチェーン間で移行できるようにするSkillOptを提案しました。 Codexで最適化されたSpreadsheetBenchスキルは、Claude Codeに適用した際に81.8点を獲得し、後者の独学スキルによる80.4点を上回りました。 4つのクロスモデル、4つのクロスツールチェーン、3つのクロスベンチマーク移行結果はすべて、ターゲットのノースキルベースラインを上回っていました。","category":"論文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"MicrosoftのSkillOptは、最適化されたエージェントスキルのアーティファクトが異なるモデルサイズやCodex、Claudeコード間で移行できることを証明しています - Aioga AIニュース","description":"Microsoftは上海交通大学、通済大学、復旦のチームと共に、単一スキル文書をテキスト空間最適化で訓練し、ターゲットモデルを凍結し、最適化されたスキルアーティファクトをモデルスケールやツールチェーン間で移行できるようにするSkillOptを提案しました。 Codexで最適化されたSpreadsheetBenchスキルは、Claude Codeに適用した際に...","url":"https://www.aioga.com/ja/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:01.695Z"},"ko":{"title":"마이크로소프트의 SkillOpt은 최적화된 에이전트 스킬 아티팩트가 서로 다른 모델 크기와 Codex 및 Claude 코드 간에 이동할 수 있음을 증명합니다","summary":"마이크로소프트는 상하이교통대학교, 퉁지, 푸단 팀과 함께 단일 기술 문서를 텍스트 공간 최적화를 통해 학습시키고, 대상 모델을 동결하며, 최적화된 기술 산출물을 모델 규모와 도구 체인 간에 이동할 수 있도록 하는 SkillOpt를 제안했습니다. Codex에서 최적화된 SpreadsheetBench 기술은 Claude Code에 적용했을 때 81.8점을 기록하며, Claude Code가 독학한 80.4점을 능가했습니다. 네 개의 크로스 모델, 네 개의 크로스 툴체인, 세 개의 크로스 벤치마크 마이그레이션 결과 모두 목표의 무기술 기준선을 초과했습니다.","category":"연구","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"마이크로소프트의 SkillOpt은 최적화된 에이전트 스킬 아티팩트가 서로 다른 모델 크기와 Codex 및 Claude 코드 간에 이동할 수 있음을 증명합니다 - Aioga AI 뉴스","description":"마이크로소프트는 상하이교통대학교, 퉁지, 푸단 팀과 함께 단일 기술 문서를 텍스트 공간 최적화를 통해 학습시키고, 대상 모델을 동결하며, 최적화된 기술 산출물을 모델 규모와 도구 체인 간에 이동할 수 있도록 하는 SkillOpt를 제안했습니다. Codex에서 최적화된 SpreadsheetBench 기술은 Claude Co...","url":"https://www.aioga.com/ko/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:10.446Z"},"es":{"title":"SkillOpt de Microsoft demuestra que los artefactos optimizados de habilidades de agentes pueden migrar entre diferentes tamaños de modelo y Codex and Claude Code","summary":"Microsoft, junto con equipos de la Universidad Jiao Tong de Shanghái, Tongji y Fudan, propuso SkillOpt, que entrena documentos de habilidad única mediante optimización del espacio de texto, congela modelos objetivo y permite que artefactos de habilidades optimizados migren entre escalas y cadenas de herramientas. La habilidad optimizada de SpreadsheetBench en Codex obtuvo 81,8 cuando se implementó en Claude Code, superando el 80,4 logrado por las habilidades autodidactas de este último. Los cuatro resultados de migración cruzados entre modelos, cuatro entre cadenas de herramientas y tres entre benchmarks estuvieron por encima del nivel base de ausencia de habilidades del objetivo.","category":"Investigación","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt de Microsoft demuestra que los artefactos optimizados de habilidades de agentes pueden migrar entre diferentes tamaños de modelo y Codex and Claude Code - Aioga Noticias de IA","description":"Microsoft, junto con equipos de la Universidad Jiao Tong de Shanghái, Tongji y Fudan, propuso SkillOpt, que entrena documentos de habilidad única mediante optimización del espacio...","url":"https://www.aioga.com/es/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:10.014Z"},"fr":{"title":"SkillOpt de Microsoft prouve que les artefacts de compétences d’agent optimisés peuvent migrer entre différentes tailles de modèles et Codex et Code Claude","summary":"Microsoft, avec des équipes de l’Université Jiao Tong de Shanghai, de Tongji et de Fudan, a proposé SkillOpt, qui entraîne des documents à compétence unique grâce à l’optimisation de l’espace textuel, fige les modèles cibles et permet de migrer des artefacts de compétences optimisés à travers les échelles et chaînes d’outils des modèles. La compétence SpreadsheetBench optimisée sur Codex a obtenu 81,8 lors de son déploiement sur Claude Code, dépassant les 80,4 obtenus par les compétences autodidactes de ce dernier. Les quatre résultats de migration croisée entre modèles, quatre inter-chaînes d’outils et trois cross-benchmarks étaient tous au-dessus du seuil de base de l’absence de compétences de la cible.","category":"Recherche","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt de Microsoft prouve que les artefacts de compétences d’agent optimisés peuvent migrer entre différentes tailles de modèles et Codex et Code Claude - Aioga Actualités IA","description":"Microsoft, avec des équipes de l’Université Jiao Tong de Shanghai, de Tongji et de Fudan, a proposé SkillOpt, qui entraîne des documents à compétence unique grâce à l’optimisation...","url":"https://www.aioga.com/fr/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:19.089Z"},"de":{"title":"Microsofts SkillOpt beweist, dass optimierte Agenten-Skill-Artefakte zwischen verschiedenen Modellgrößen und Codex sowie Claude Code wechseln können","summary":"Microsoft schlug zusammen mit Teams der Shanghai Jiao Tong University, Tongji und Fudan SkillOpt vor, das Single-Skill-Dokumente durch Textraumoptimierung trainiert, Zielmodelle einfriert und optimierte Skill-Artefakte über Modellskalen und Toolchains hinweg migriert. Die auf Codex optimierte SpreadsheetBench-Fähigkeit erreichte 81,8 bei der Anwendung auf Claude Code und übertraf damit die 80,4, die durch dessen selbstgelernte Fähigkeiten erreicht wurde. Alle vier Cross-Model-, 4-Cross-Toolchain- und drei Cross-Benchmark-Migrationsergebnisse lagen über dem No-Skills-Baseline des Ziels.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsofts SkillOpt beweist, dass optimierte Agenten-Skill-Artefakte zwischen verschiedenen Modellgrößen und Codex sowie Claude Code wechseln können - Aioga KI-News","description":"Microsoft schlug zusammen mit Teams der Shanghai Jiao Tong University, Tongji und Fudan SkillOpt vor, das Single-Skill-Dokumente durch Textraumoptimierung trainiert, Zielmodelle ei...","url":"https://www.aioga.com/de/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:19.112Z"},"pt-BR":{"title":"O SkillOpt da Microsoft prova que artefatos otimizados de habilidades de agentes podem migrar entre diferentes tamanhos de modelo e Codex e Claude Code","summary":"A Microsoft, junto com equipes da Universidade Jiao Tong de Xangai, Tongji e Fudan, propôs o SkillOpt, que treina documentos de habilidade única por meio de otimização de espaço de texto, congela modelos-alvo e permite que artefatos de habilidades otimizados migrem entre escalas e cadeias de ferramentas de modelos. A habilidade otimizada do SpreadsheetBench no Codex teve 81,8 quando aplicada ao Claude Code, superando os 80,4 alcançados pelas habilidades autodidatas deste último. Todos os quatro resultados de migração cross-model, cross-toolchain e três cross-benchmark estavam acima do critério de ausência de habilidades do alvo.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"O SkillOpt da Microsoft prova que artefatos otimizados de habilidades de agentes podem migrar entre diferentes tamanhos de modelo e Codex e Claude Code - Aioga Notícias de IA","description":"A Microsoft, junto com equipes da Universidade Jiao Tong de Xangai, Tongji e Fudan, propôs o SkillOpt, que treina documentos de habilidade única por meio de otimização de espaço de...","url":"https://www.aioga.com/pt-BR/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:27.741Z"},"ru":{"title":"SkillOpt от Microsoft доказывает, что оптимизированные артефакты навыков агентов могут мигрировать между разными размерами моделей и кодом Codex и Claude","summary":"Microsoft вместе с командами из Шанхайского университета Цзяо Тун, Тунцзи и Фудань предложили SkillOpt, который обучает документы по одному навыку через оптимизацию текстового пространства, замораживает целевые модели и позволяет мигрировать оптимизированные артефакты навыков между масштабами моделей и цепочками инструментов. Навык SpreadsheetBench, оптимизированный на Codex, получил 81,8 балла при развертывании в Claude Code, превзойдя 80,4, достигнутый самообученными навыками последнего. Все четыре результата кросс-модели, четыре кросс-инструментчейна и три кросс-бенчмарновенных результата миграции были выше базового уровня отсутствия навыков для целевой цели.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt от Microsoft доказывает, что оптимизированные артефакты навыков агентов могут мигрировать между разными размерами моделей и кодом Codex и Claude - Aioga Новости ИИ","description":"Microsoft вместе с командами из Шанхайского университета Цзяо Тун, Тунцзи и Фудань предложили SkillOpt, который обучает документы по одному навыку через оптимизацию текстового прос...","url":"https://www.aioga.com/ru/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:27.917Z"},"ar":{"title":"يثبت برنامج SkillOpt من مايكروسوفت أن تشوهات مهارات الوكلاء المحسنة يمكن أن تنتقل بين أحجام النماذج المختلفة وكودكس وكلود كود","summary":"اقترحت مايكروسوفت، مع فرق من جامعة شنغهاي جياو تونغ، وتونغجي، وفودان، تطبيق SkillOpt، الذي يدرب مستندات المهارات الفردية من خلال تحسين مساحة النصوص، ويجمد نماذج الأهداف، وتمكين التشويهات المحسنة من المهارات من التنقل عبر مقاييس النماذج وسلاسل الأدوات. حصلت مهارة SpreadsheetBench المحسنة على Codex على 81.8 عند نشرها في كود كلود، متجاوزة 80.4 التي حققها مهارات الأخير الذاتي. جميع نتائج الهجرة عبر النماذج الأربعة، وأربع عبر سلسلة الأدوات، وثلاث نتائج عبر المعايير كانت أعلى من خط الأساس الخالي من المهارات للهدف.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"يثبت برنامج SkillOpt من مايكروسوفت أن تشوهات مهارات الوكلاء المحسنة يمكن أن تنتقل بين أحجام النماذج المختلفة وكودكس وكلود كود - Aioga أخبار الذكاء الاصطناعي","description":"اقترحت مايكروسوفت، مع فرق من جامعة شنغهاي جياو تونغ، وتونغجي، وفودان، تطبيق SkillOpt، الذي يدرب مستندات المهارات الفردية من خلال تحسين مساحة النصوص، ويجمد نماذج الأهداف، وتمكين الت...","url":"https://www.aioga.com/ar/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:35.012Z"},"hi":{"title":"माइक्रोसॉफ्ट का स्किलऑप्ट साबित करता है कि अनुकूलित एजेंट कौशल कलाकृतियां विभिन्न मॉडल आकारों और कोडेक्स और क्लाउड कोड के बीच माइग्रेट कर सकती हैं","summary":"माइक्रोसॉफ्ट ने शंघाई जिओ टोंग विश्वविद्यालय, टोंगजी और फुदान की टीमों के साथ मिलकर स्किलऑप्ट का प्रस्ताव रखा, जो टेक्स्ट स्पेस ऑप्टिमाइज़ेशन के माध्यम से एकल-कौशल दस्तावेजों को प्रशिक्षित करता है, लक्ष्य मॉडल को फ्रीज करता है, और अनुकूलित कौशल कलाकृतियों को मॉडल स्केल और टूलचेन में माइग्रेट करने में सक्षम बनाता है। कोडेक्स पर अनुकूलित स्प्रेडशीटबेंच कौशल ने क्लाउड कोड में तैनात होने पर 81.8 स्कोर किया, जो बाद के स्व-प्रशिक्षित कौशल द्वारा हासिल किए गए 80.4 को पार कर गया। सभी चार क्रॉस-मॉडल, चार क्रॉस-टूलचेन, और तीन क्रॉस-बेंचमार्क माइग्रेशन परिणाम लक्ष्य के नो-स्किल्स बेसलाइन से ऊपर थे।","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"माइक्रोसॉफ्ट का स्किलऑप्ट साबित करता है कि अनुकूलित एजेंट कौशल कलाकृतियां विभिन्न मॉडल आकारों और कोडेक्स और क्लाउड कोड के बीच माइग्रेट कर सकती हैं - Aioga AI समाचार","description":"माइक्रोसॉफ्ट ने शंघाई जिओ टोंग विश्वविद्यालय, टोंगजी और फुदान की टीमों के साथ मिलकर स्किलऑप्ट का प्रस्ताव रखा, जो टेक्स्ट स्पेस ऑप्टिमाइज़ेशन के माध्यम से एकल-कौशल दस्तावेजों को प्...","url":"https://www.aioga.com/hi/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:35.898Z"},"it":{"title":"SkillOpt di Microsoft dimostra che artefatti di abilità degli agenti ottimizzati possono migrare tra diverse dimensioni di modello e Codex e Claude Code","summary":"Microsoft, insieme a team della Shanghai Jiao Tong University, Tongji e Fudan, ha proposto SkillOpt, che addestra documenti a singola competenza tramite ottimizzazione dello spazio testuale, congela i modelli target e consente la migrazione ottimizzata degli artefatti delle abilità tra scale e toolchain di modello. La competenza SpreadsheetBench ottimizzata su Codex ha ottenuto 81,8 quando è stata applicata a Claude Code, superando gli 80,4 ottenuti dalle competenze autodidattiche di quest'ultimo. Tutti e quattro i risultati di migrazione cross-model, cross-toolchain e tre cross-benchmark erano al di sotto del target di assenza di competenze.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt di Microsoft dimostra che artefatti di abilità degli agenti ottimizzati possono migrare tra diverse dimensioni di modello e Codex e Claude Code - Aioga Notizie IA","description":"Microsoft, insieme a team della Shanghai Jiao Tong University, Tongji e Fudan, ha proposto SkillOpt, che addestra documenti a singola competenza tramite ottimizzazione dello spazio...","url":"https://www.aioga.com/it/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:44.759Z"},"nl":{"title":"Microsoft's SkillOpt bewijst dat geoptimaliseerde agentvaardigheidsartefacten kunnen migreren tussen verschillende modelgroottes en Codex en Claude Code","summary":"Microsoft stelde samen met teams van Shanghai Jiao Tong University, Tongji en Fudan SkillOpt voor, dat single-skill documenten traint via tekstruimte-optimalisatie, doelmodellen bevriest en geoptimaliseerde vaardigheidsartefacten over modelschalen en toolchains laat migreren. De SpreadsheetBench-vaardigheid die op Codex was geoptimaliseerd, scoorde 81,8 bij implementatie op Claude Code, waarmee hij de 80,4 die door de zelfgetrainde vaardigheden van laatstgenoemde werd bereikt. Alle vier cross-model, vier cross-toolchain en drie cross-benchmark migratieresultaten lagen boven de geen-skills baseline van het doelwit.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsoft's SkillOpt bewijst dat geoptimaliseerde agentvaardigheidsartefacten kunnen migreren tussen verschillende modelgroottes en Codex en Claude Code - Aioga AI-nieuws","description":"Microsoft stelde samen met teams van Shanghai Jiao Tong University, Tongji en Fudan SkillOpt voor, dat single-skill documenten traint via tekstruimte-optimalisatie, doelmodellen be...","url":"https://www.aioga.com/nl/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:44.601Z"},"tr":{"title":"Microsoft'un SkillOpt'u, optimize edilmiş ajan becerisi artefaktlarının farklı model boyutları ile Codex ile Claude Code arasında geçiş yapabildiğini kanıtlıyor","summary":"Microsoft, Shanghai Jiao Tong Üniversitesi, Tongji ve Fudan ekipleriyle birlikte, metin alanı optimizasyonuyla tek yetenekli belgeleri eğiten, hedef modelleri donduran ve optimize edilmiş beceri eserlerinin model ölçekleri ve araç zincirleri arasında taşınmasını sağlayan SkillOpt'u önerdi. Codex'te optimize edilen SpreadsheetBench becerisi, Claude Code'a uygulandığında 81.8 puan aldı ve kendi kendine eğittiği becerilerin elde ettiği 80.4'ü geride bıraktı. Dört çapraz model, dört çapraz araç zinciri ve üç çapraz ölçüt göç sonuçları hedefin beceri eksikliği tabanının üzerindeydi.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsoft'un SkillOpt'u, optimize edilmiş ajan becerisi artefaktlarının farklı model boyutları ile Codex ile Claude Code arasında geçiş yapabildiğini kanıtlıyor - Aioga AI Haberleri","description":"Microsoft, Shanghai Jiao Tong Üniversitesi, Tongji ve Fudan ekipleriyle birlikte, metin alanı optimizasyonuyla tek yetenekli belgeleri eğiten, hedef modelleri donduran ve optimize...","url":"https://www.aioga.com/tr/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:53.644Z"},"vi":{"title":"SkillOpt của Microsoft chứng minh rằng các artifact kỹ năng agent được tối ưu hóa có thể di chuyển giữa các kích thước mô hình khác nhau và Codex và Claude Code","summary":"Microsoft, cùng với các nhóm từ Đại học Giao thông Thượng Hải, Đồng Cơ và Phúc Đán, đã đề xuất SkillOpt, giúp huấn luyện tài liệu một kỹ năng thông qua tối ưu hóa không gian văn bản, đóng băng các mô hình mục tiêu và cho phép các hiện vật kỹ năng tối ưu di chuyển qua các quy mô và chuỗi công cụ. Kỹ năng SpreadsheetBench tối ưu hóa trên Codex đạt 81.8 điểm khi triển khai cho Claude Code, vượt qua mức 80.4 của kỹ năng tự luyện của Code. Cả bốn kết quả di chuyển đa mô hình, bốn chuỗi công cụ chéo và ba kết quả di chuyển chéo đều vượt mức cơ bản không có kỹ năng của mục tiêu.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt của Microsoft chứng minh rằng các artifact kỹ năng agent được tối ưu hóa có thể di chuyển giữa các kích thước mô hình khác nhau và Codex và Claude Code - Tin tức AI Aioga","description":"Microsoft, cùng với các nhóm từ Đại học Giao thông Thượng Hải, Đồng Cơ và Phúc Đán, đã đề xuất SkillOpt, giúp huấn luyện tài liệu một kỹ năng thông qua tối ưu hóa không gian văn bả...","url":"https://www.aioga.com/vi/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:25:52.961Z"},"id":{"title":"SkillOpt dari Microsoft membuktikan bahwa artefak skill agen yang dioptimalkan dapat bermigrasi antara ukuran model yang berbeda dan Codex serta Claude Code","summary":"Microsoft, bersama tim dari Shanghai Jiao Tong University, Tongji, dan Fudan, mengusulkan SkillOpt, yang melatih dokumen keterampilan tunggal melalui optimasi ruang teks, membekukan model target, dan memungkinkan artefak keterampilan yang dioptimalkan untuk bermigrasi melintasi skala model dan rantai alat. Keterampilan SpreadsheetBench yang dioptimalkan pada Codex mendapatkan skor 81,8 saat diterapkan pada Claude Code, melampaui 80,4 yang dicapai oleh keterampilan mandiri Claude Codex. Keempat hasil migrasi lintas model, empat lintas rantai alat, dan tiga hasil migrasi lintas benchmark berada di atas batas dasar tanpa keterampilan pada target.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt dari Microsoft membuktikan bahwa artefak skill agen yang dioptimalkan dapat bermigrasi antara ukuran model yang berbeda dan Codex serta Claude Code - Berita AI Aioga","description":"Microsoft, bersama tim dari Shanghai Jiao Tong University, Tongji, dan Fudan, mengusulkan SkillOpt, yang melatih dokumen keterampilan tunggal melalui optimasi ruang teks, membekuka...","url":"https://www.aioga.com/id/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:26:02.312Z"},"th":{"title":"SkillOpt ของ Microsoft พิสูจน์ว่าอาร์ติแฟกต์ทักษะเอเจนต์ที่ได้รับการปรับแต่งแล้วสามารถย้ายระหว่างขนาดโมเดลต่าง ๆ และโค้ด Codex กับ Claude ได้","summary":"Microsoft ร่วมกับทีมจากมหาวิทยาลัย Shanghai Jiao Tong, Tongji และ Fudan ได้เสนอ SkillOpt ซึ่งฝึกเอกสารทักษะเดียวผ่านการปรับแต่งพื้นที่ข้อความ หยุดโมเดลเป้าหมาย และช่วยให้อาร์ติแฟกต์ทักษะที่ได้รับการปรับแต่งสามารถย้ายข้ามขนาดและเครื่องมือต่าง ๆ ได้ ทักษะ SpreadsheetBench ที่ปรับแต่งใน Codex ได้คะแนน 81.8 เมื่อนําไปใช้กับ Claude Code แซงหน้า 80.4 ที่ Claude Code ฝึกด้วยตัวเอง ผลลัพธ์การย้ายข้อมูลข้ามโมเดลทั้งสี่แบบ ข้ามเครื่องมือสี่ครั้ง และผลลัพธ์ข้ามเกณฑ์มาตรฐานสามครั้งอยู่เหนือค่าพื้นฐานไม่มีทักษะของเป้าหมาย","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"SkillOpt ของ Microsoft พิสูจน์ว่าอาร์ติแฟกต์ทักษะเอเจนต์ที่ได้รับการปรับแต่งแล้วสามารถย้ายระหว่างขนาดโมเดลต่าง ๆ และโค้ด Codex กับ Claude ได้ - ข่าว AI Aioga","description":"Microsoft ร่วมกับทีมจากมหาวิทยาลัย Shanghai Jiao Tong, Tongji และ Fudan ได้เสนอ SkillOpt ซึ่งฝึกเอกสารทักษะเดียวผ่านการปรับแต่งพื้นที่ข้อความ หยุดโมเดลเป้าหมาย และช่วยให้อาร์ติแฟกต...","url":"https://www.aioga.com/th/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:26:01.250Z"},"pl":{"title":"Microsoft SkillOpt udowadnia, że zoptymalizowane artefakty umiejętności agenta mogą migrować między różnymi rozmiarami modeli oraz kodem Codex i Claude","summary":"Microsoft wraz z zespołami z Shanghai Jiao Tong University, Tongji i Fudan zaproponował SkillOpt, który trenuje dokumenty z pojedynczą umiejętnością poprzez optymalizację przestrzeni tekstowej, zamraża modele docelowe i umożliwia migrację zoptymalizowanych artefaktów umiejętności pomiędzy skalami modeli i narzędziami. Umiejętność SpreadsheetBench zoptymalizowana w Codex uzyskała 81,8 po wdrożeniu w Claude Code, przewyższając 80,4 uzyskane dzięki samodzielnym umiejętnościom tego drugiego. Wszystkie cztery wyniki migracji z różnych modeli, cztery z łańcucha narzędzi i trzy z benchmarków były powyżej poziomu wyjściowego braku umiejętności dla celu.","category":"论文研究","source":"MarkTechPost（RSS）","aggregationSource":"MarkTechPost（RSS）","pageTitle":"Microsoft SkillOpt udowadnia, że zoptymalizowane artefakty umiejętności agenta mogą migrować między różnymi rozmiarami modeli oraz kodem Codex i Claude - Aioga Wiadomości AI","description":"Microsoft wraz z zespołami z Shanghai Jiao Tong University, Tongji i Fudan zaproponował SkillOpt, który trenuje dokumenty z pojedynczą umiejętnością poprzez optymalizację przestrze...","url":"https://www.aioga.com/pl/news/cmsgsgz530fluro5qu0vnw8s0/","contentTranslated":true,"sourceHash":"9a8ece1a8ea6d456","translatedAt":"2026-08-06T01:26:10.982Z"}}}}