The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to reference documentation and SDK source code from verified vendor domains and repositories, such as cloud.tencent.com, docs.cloudbase.net, and github.com/TencentCloud. These downloads are used neutrally to discover and verify API schemas and parameter requirements.\n- [COMMAND_EXECUTION]: The skill utilizes the callCloudApi tool to execute control-plane operations on various Tencent Cloud services. It provides specific sequences of commands, called 'recipes', to perform tasks like configuring alerts for database instances, ensuring that all execution follows proven, documented workflows.\n- [PRIVILEGE_ESCALATION]: The skill manages agent permissions by providing users with one-click authorization links to the official Tencent Cloud console. This allows users to manually grant specific pre-defined Access Management (CAM) policies (e.g., QcloudMonitorFullAccess) to the agent's service role, maintaining explicit user control over permission changes.\n- [INDIRECT_PROMPT_INJECTION]: The skill implements a strategy to mitigate risks associated with processing user-provided API parameters. It requires the agent to verify all action names and parameter shapes against official documentation or SDK source code before execution. The evidence chain is as follows: \n * Ingestion points: User input provided for API parameters in SKILL.md and references/recipes/pg-storage-alarm.md.\n * Boundary markers: The structured nature of the callCloudApi parameters. \n * Capability inventory: The use of callCloudApi for cloud resource management as documented in references/calling-methods.md.\n * Sanitization: The mandatory requirement to cross-reference input against official vendor SDK models to prevent the agent from being misled by injected instructions in user-controlled data.

cloud-api-operations