下载地址
https://www.zhengdianxue.com/wp-content/uploads/skills/yahuixun--cloud-manage-skill.zip将提示词发送给你的 AI 安装该 skills
请根据 https://skillhub.cn/install/skillhub.md,安装 @yahuixun/cloud-manage-skill。Use the local multi-cloud management gateway for cloud task planning, audit retrieval, and future CLI/MCP workflows across providers.
阅读全文 (展开查看完整说明)
概述
Cloud Manage Skill
Use this local service as the audited boundary for cloud management Agent tasks.
The target control path is CLI/MCP through the Docker-hosted gateway. Tencent Cloud is the first provider track; AWS, Azure, and Google Cloud are planned provider tracks. Browser automation is outside this gateway and should use the agent's browser or Playwright Skill.
Prerequisite
Confirm the service is running:
curl http://127.0.0.1:8000/healthContinue only when the response is {"status":"ok"}. If a visible console session or QR-code login is needed, use the agent's browser or Playwright Skill.
Allowed Current Actions
- Create a Tencent Cloud dry-run plan and receive a plan fingerprint:
POST /tasks/tencent/plan- Record human confirmation for a completed dry-run task:
POST /tasks/confirm- Execute Tencent Cloud operations:
POST /tasks/tencent/execute- Review a task, its timestamps, and logs:
GET /tasks/{task_id}- Review recent tasks:
GET /tasksSafety Workflow
For Tencent Cloud operations, always start with POST /tasks/tencent/plan to generate a dry-run record and fingerprint.
POST /tasks/tencent/execute enforces these rules:
- Read-only operations can execute directly after policy classification.
- Dangerous operations must provide both
dry_run_task_idandconfirmation_id. - The execution request must match the approved dry-run plan fingerprint exactly.
- Dangerous operations are currently blocked after validation in this phase.
Do not implement or invoke an action that mutates cloud resources, account settings, permissions, security controls, or billing unless:
- A dry-run task records the exact intended change.
- The human reviews and explicitly confirms the planned change.
- The execution is separately logged with timestamps and command evidence.
Never request, store, or infer passwords.
For future CLI execution, credentials must come from runtime environment variables, Docker secrets, or provider-approved credential files. Never place credentials in prompts, commands, task logs, Docker images, or repository files.
Response Handling
Report the task ID, action status, and visible result when present. When a task fails, surface its logged error rather than attempting broader permissions.