论文提出了 SoftSkill，能把复杂技能压缩成几个虚拟向量，用 Qwen3.5-4B 测试，数学题直接涨了 42 个点，比 SkillOpt 还强。
SoftSkill 提出一种方法，将智能体的自然语言技能（Markdown 文件）转化为紧凑的连续上下文对象，通过可训练的软增量进行微调，而基础模型保持不变。在单轮设置下，Qwen3.5-4B 使用长度为 32 的 SoftSkill 前缀相比无技能提示，在 SearchQA 上提升 8.3 点，LiveMath 提升 42.1 点，DocVQA 提升 1.3 点。与 SkillOpt 相比，SoftSkill 在 SearchQA 上准确率提升 5.2 点，LiveMath 提升 12.5 点，同时将数百到数千个 Markdown 技能令牌替换为几个虚拟令牌。论文还探讨了智能体执行作为更难边界情况下的表现。
SoftSkill: Behavioral Compression for Contextual Adaptation
Agent skills are commonly deployed as natural-language Markdown files that encode answer policies, evidence-use habits, and task procedures. These files are readable and portable, but they are consumed indirectly: for each task instance, a frozen language model must translate a long textual artifact into generation-time behavior. This paper asks whether a natural-language skill can instead initialize a compact continuous context object, refined by a trainable soft delta while the base model remains frozen. We propose SoftSkill, a frozen-backbone method that tunes such soft skills with next-token prediction and deploys them as latent behavioral priors at inference time. In our main single-round setting, a length-32 SoftSkill prefix on Qwen3.5-4B improves over no-skill prompting by 8.3 points on SearchQA, 42.1 points on LiveMath, and 1.3 points on DocVQA. Relative to SkillOpt, SoftSkill improves accuracy by 5.2 points on SearchQA and 12.5 points on LiveMath, while replacing hundreds to thousands of Markdown skill tokens with a few virtual tokens. We further study agentic execution as a harder boundary case, where sparse trajectory imitation provides useful signal but does not yet robustly compress long-horizon procedural behavior. More broadly, the results suggest that some task skills are better treated not as additional Markdown to be reinterpreted at inference time, but as compact latent controls over how a frozen model enters the task.