问题描述
当前 browser-skill 对 hover 触发型菜单的端到端支持不完整:CLI/daemon/extension 缺少稳定的 hover 动作链路,VOM probe 虽然能发现 hover 后出现的菜单内容,但初始 observe 有时不会把 hover 引导信息标注到实际可操作的 handle 上,导致 agent 可能直接 click 触发器,而不是先 hover 再操作展开菜单。
需要从机制上补齐 hover RPC 与 VOM hover surface 映射,使 agent 能稳定完成:
observe 看到可 hover handle 和菜单摘要 → hover @ref 展开并保持菜单 → 再次 observe 获取菜单项 refs → click 目标菜单项。
Problem Description
Browser-skill currently has incomplete end-to-end support for hover-triggered menus. The CLI/daemon/extension stack lacks a stable hover action path, and although VOM probes can discover menu content that appears after hover, the initial observe output may fail to attach hover guidance to the actual actionable handle. As a result, the agent may incorrectly click the trigger directly instead of hovering first and then interacting with the expanded menu.
We need a mechanism-level fix for both the hover RPC path and VOM hover surface mapping, so the agent can reliably complete this flow:
observe shows a hoverable handle and menu summary → hover @ref expands and keeps the menu open → another observe returns refs for the menu items → click the target menu item.
问题描述
当前 browser-skill 对 hover 触发型菜单的端到端支持不完整:CLI/daemon/extension 缺少稳定的 hover 动作链路,VOM probe 虽然能发现 hover 后出现的菜单内容,但初始 observe 有时不会把 hover 引导信息标注到实际可操作的 handle 上,导致 agent 可能直接 click 触发器,而不是先 hover 再操作展开菜单。
需要从机制上补齐 hover RPC 与 VOM hover surface 映射,使 agent 能稳定完成:
observe 看到可 hover handle 和菜单摘要 → hover @ref 展开并保持菜单 → 再次 observe 获取菜单项 refs → click 目标菜单项。
Problem Description
Browser-skill currently has incomplete end-to-end support for hover-triggered menus. The CLI/daemon/extension stack lacks a stable hover action path, and although VOM probes can discover menu content that appears after hover, the initial observe output may fail to attach hover guidance to the actual actionable handle. As a result, the agent may incorrectly click the trigger directly instead of hovering first and then interacting with the expanded menu.
We need a mechanism-level fix for both the hover RPC path and VOM hover surface mapping, so the agent can reliably complete this flow:
observe shows a hoverable handle and menu summary → hover @ref expands and keeps the menu open → another observe returns refs for the menu items → click the target menu item.