메뉴
Automate web pages using Playwright in this repo. Use when Codex needs to build or extend RPA scripts, examples, or core utilities that launch browsers, navigate, interact with elements, extract data, or orchestrate flows.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | rpa-web-automation |
| description | Automate web pages using Playwright in this repo. Use when Codex needs to build or extend RPA scripts, examples, or core utilities that launch browsers, navigate, interact with elements, extract data, or orchestrate flows. |
这是一份给 Codex/多 AI 协作开发 用的 SKILL.md（指导“怎么在这个 repo 里工作”），不是运行时的业务 skill（不是 skills/* 那种）。
目标：
只做必要步骤，逐层加复杂度：
integrations/rpaskill_ts/skills/<skill>/ + runtime/integrations/rpaskill_ts/src/：核心实现（浏览器/导航/元素/提取/流程）integrations/rpaskill_ts/src/skills/：搜索类模块（webSearch、adaptiveSearch 等）integrations/rpaskill_ts/cli/run.mjs：稳定 CLI runner（给 Python 平台调用）runtime/：engine/logger/config + outputs 规范 + events/memory/artifactsskills/：平台一等公民 skills（也可以做 TS 的薄封装 skill）建议用平台的共享 venv（避免每个 skill 各装各的）：
python setup_platform.py --skip-browsers浏览器二进制默认规划在：
runtime/deps/playwright_browsers/runtime/engine.py 统一设置 PLAYWRIGHT_BROWSERS_PATH如果网络环境导致下载浏览器失败：
channel=msedge / channel=chrome（最稳定）在 integrations/rpaskill_ts/：
npm install关键原则：默认只写自己的 agent 目录，共享目录只让一个“协调者 agent”写。
运行时目录（每个 run）：
outputs/<skill>/<run_id>/
shared/
agents/<agent_id>/
work/
events.jsonl
memory.jsonl
index.jsonl
result.json
建议写法：
outputs/.../agents/<agent_id>/work/（临时文件/中间结果/草稿）shared/memory.jsonl、shared/events.jsonlevents.jsonl：结构化事件（event/level/data/paths）memory.jsonl：短“事实/决策”（大内容写 artifact 文件，再在 memory 里引用路径）index.jsonl：产物索引（路径/hash/大小）// TS 端实现
async function checkForHumanInterventionNeeded(page): Promise<boolean> {
// 检查是否出现登录页、验证码等需要人工干预的场景
const hasLoginForm = await page.locator('#login-form').isVisible().catch(() => false);
const hasCaptcha = await page.locator('.captcha').isVisible().catch(() => false);
return hasLoginForm || hasCaptcha;
}
async function requestHumanIntervention(page, taskDescription): Promise<boolean> {
console.log(`[HUMAN INTERVENTION NEEDED] ${taskDescription}`);
console.log('Please complete the task in the browser window.');
console.log('Press Enter when done...');
// 等待用户输入
await new Promise(resolve => {
process.stdin.resume();
process.stdin.once('data', () => {
process.stdin.pause();
resolve(true);
});
});
return true;
}
# Python 端实现
def run_with_human_intervention(ctx, rpa_task):
"""运行 RPA 任务，遇到需要人工干预的场景时暂停等待"""
try:
result = rpa_task()
return result
except HumanInterventionRequired as e:
print(f"[需要人工干预] {e.message}")
print("请在浏览器中完成操作，完成后按 Enter 继续...")
input()
# 继续执行任务
return rpa_task()
waitForLoadState / element visible，再 click/typesleep：用 retry + timeout 提升稳定性integrations/rpaskill_ts/src/core/runtime/rpa/ 或 runtime/common/integrations/rpaskill_ts/examples/skills/<skill>/codex-skills/search-enhancer/SKILL.md