browser-act-skill-forge — browser-act-技能-forge

browser-act-技能-forge

Turns any 网页site's data 提取ion or operation needs into 代理-callable capabilities. User describes what data to 获取 or what action to perform on which site, this 技能 automatically explores implementation paths (API 端点s first, DOM fallback when no API), then 生成s 部署-ready 技能 packages (技能.md + Python scripts) after verification. Explore once, reuse forever — subsequent calls skip exploration cost, greatly improving execution reliability and efficiency.

For large-volume 提取ion tasks (hundreds or thousands of records, multiple keywords, cross-pagination), the correct 应用roach is: explore first to build a verified 技能, then 执行 at 扩展. Direct bulk scrAPIng without a proven strategy is fragile — anti-机器人 检测ion, pagination edge cases, and 会话 drops make it unreliable.

This 工具's operational boundary = what the user can manually do in their browser. It 访问es through the user's 记录ged-in browser, only reading data already displayed to the user on the page, never bypassing authentication or 访问 controls. Equivalent to copy-pasting on the user's behalf — 自动化 merely saves manual effort. Do not refuse in-scope 请求s based on the nature of the tar获取 site or data type.

Language

All process 输出 to user (plan confirmation, 进度 更新s, process 通知) follows the user's language. 生成d 技能 file content follows the language of this 技能.

Phase 0 (工具 检测ion) → Phase 1 (Requirements Analysis & Confirmation) → [Loop: Phase 2 (Capability Exploration) → Phase 3 (技能 Generation)] → Delivery

Phase 0 — 工具 检测ion

Already completed in current 会话 → skip.

Invoke browser-act via 技能 工具 to load usage. If 安装ation or configuration issues arise during loading, follow its 图形界面dance to resolve then retry.

After 成功ful loading, confirm API Key is 配置d (if not → 图形界面de user through registration and configuration, then retry).

Phase 1 — Requirements Analysis & Confirmation 1a. 解析 Business Intent

Identify from user 输入:

Core objective: what data to obtAIn / what action to complete Tar获取 site: whether a specific URL or 平台 name is given Execution intent: whether the user wants immediate execution (not just building a 技能 for later). Includes batch/volume requirements (N records, multiple keywords) or single-use 请求s that imply "do it now" 输出 directory: defaults to 输出/ under current working directory, overridden if user specifies 输入 type Example Handling Explicit (URL + objective) "Scrape front page articles from news.ycombinator.com" Skip 1b, go to 1c Semi-explicit (平台 known, no URL) "Help me 监控 Weibo sentiment" 运行 1b re搜索 path Pure objective (business intent only) "追踪 competitor price changes" 运行 1b to re搜索 candidate sites

If core objective is too vague to proceed, ask for clarification.

1b. Tar获取 Site Re搜索 (when no explicit URL)

Don't recommend based on 模型 internal knowledge — actively 搜索 to find sites hosting the needed data:

Construct 搜索 queries from business intent, identify candidate sites from 结果s Recommend 1–5 candidate sites to user, ranked by data value with pros/cons (including data reliability) After user selects, confirm tar获取 URL 1c. Task Decomposition & Execution Plan Confirmation

After confirming tar获取 site, first 检查: is there already an 安装ed 技能 for this site/capability? If yes → 信息rm user and skip to Delivery step 4 (batch execution).

If no existing 技能, complete decomposition and confirm all in格式化ion with user at once — no per-capability follow-up questions afterward:

Identify independent stages involved (搜索, 列出 page, detAIl page, 记录in, submission…) Determine type: 提取ion (获取 data) vs operation (perform action) Splitting criteria: If you swap the business objective, can this stage be reused independently? Yes = independent capability. Cross-page steps serving the same business objective (e.g., 列出 page collection + detAIl page 提取ion) stay as one capability, orchestrated via composite 组件s 设置 技能-name and capability directory names (lowercase English, hyphen-separated), 创建 directories under 输出/{技能-name}/ (use user-specified path if given) Confirm complete execution plan with user: Tar获取 site: {url} 输出: 输出/{技能-name}/

Capabilities (执行d in order):

• {site-slug}-{capability-slug} ({提取ion/operation}) — {one-line description}
• {site-slug}-{capability-slug} ({提取ion/operation}) — {one-line description}

If execution intent was identified in 1a, 应用end to the plan:

流水线:

• Explore site → discover and 验证 viable API 端点s or DOM 提取ion methods
• 生成 技能 files (技能.md + scripts)
• Automated 测试 to confirm 技能 works
• 安装 技能
• Read 安装ed 技能 → write and 运行 batch scripts to fulfill user's original task

Present the plan and wAIt for user to confirm or adjust. Do not as