ask-ai1.2

遇难题时自动调用豆包/千问/Kimi/DeepSeek AI，按优先级轮询获取具体解决方案并执行方案。

Security Scan

The skill's stated purpose (asking other web AIs for help) matches its instructions, but it autonomously uses a persistent browser profile (user cookies), writes AI replies to memory, and forces execution without per-use confirmation — behaviors that are disproportionate and risk exposing credentials or enabling unexpected actions.

Purpose & Capability

Claim: delegate difficult problems to external AIs and obtain/execute solutions. The runbook uses a browser automation flow to access chat UIs — that is coherent. However the requirement to 'execute the obtained solution' and to persist cookies across sessions implies capabilities (performing actions, long-lived access to many sites) beyond a pure query/relay skill and are not declared in metadata.

Instruction Scope

SKILL.md instructs the agent to open a persistent browser profile (--browser-profile openclaw), use the user's logged-in sessions to interact with third‑party AI websites, poll for replies, save replies to memory, and then execute the returned solution without asking the user. It also mandates strict automation rules (no thinking after opening browser) and automatic AI switching. This grants broad discretion to access external web services and reuse user session cookies without explicit per-invocation consent.

Install Mechanism

Instruction-only skill with no install spec or code files — lowest install risk. There is no download/extract or third-party package installation specified.

Credentials

No env vars are declared, but the skill requires using a persistent browser profile that stores cookies and session state. That profile can act like credentials for many services (not just the target AI sites). Persisting and reusing a shared profile named 'openclaw' is disproportionate to a simple query task and could expose unrelated account sessions or site access.

Persistence & Privilege

While always:false, the skill requires and enforces persistent browser cookies and writing AI replies to memory. Combined with autonomous invocation rules (triggering without asking the user under many conditions) and a mandate to execute returned plans, this increases the agent's effective privilege and blast radius if misused or buggy.

What to consider before installing

This skill will automatically open a persistent browser profile and use whatever is logged in there to talk to third‑party AI websites, save their replies to agent memory, and then execute the suggested solutions without asking you each time. Before installing, consider: (1) Do you trust the agent to use your logged‑in accounts and cookies for these external services? (2) Use an isolated browser profile (not a shared/global one) or disable cookie persistence if you install it. (3) Disable or audit memory writes so sensitive prompts/responses aren’t stored long‑term. (4) Require a manual confirmation step before the skill executes any plan it receives. (5) Monitor activity logs and restrict the agent’s tooling/command execution permissions. If you cannot enforce these mitigations, treat this skill as high risk and avoid installing it.

执行检查清单

• 【前置】组织问题（在打开浏览器前完成，不等待）
• 步骤1：打开浏览器（带 --browser-profile openclaw）
• 步骤2：snapshot确认页面加载 + 检查登录状态
• 步骤3：直接输入已组织的问题并发送
• 步骤4：轮询等待响应（最多7次，每次间隔10秒）
• 步骤5：获取回答，关闭标签页
• 步骤6：评估回答有效性 → 决定执行/追问/切换

---

注意事项

1. 禁止白嫖：使用用户的AI账号对话，已登录状态不需重新登录
2. 有记录：把AI的回复和执行过程保存到memory
3. 闭环：拿到方案后必须执行，不能只给答案
4. 复盘：每次执行完都要总结改进点
5. 强制约束：所有输出标识必须输出，未输出禁止进入下一步
6. 切换AI：当前AI失败后自动切换下一个，不询问用户
7. 告知用户：所有AI都失败或遇到验证码时，明确告知状态
8. Cookie持久化：使用 --browser-profile openclaw（持久化托管浏览器），这是最高优先级规则
9. 轮询间隔：统一使用10秒/次，不可在执行中更改
10. 前置组织问题：触发后立即在脑中组织问题，浏览器打开后直接提交，禁止在浏览器内等待或组织问题