The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes user requirements to perform management actions on cloud infrastructure.
Ingestion point: The skill ingests user specifications and project source files (e.g., SKILL.md, references/spec-workflow/SKILL.md).
Capability inventory: Utilizes MCP tools with high privileges, including manageFunctions (deploy code), managePgDatabase (execute SQL), and managePermissions (update access control policies).
Boundary markers: Implements mandatory safety protocols ('Change Safety Protocol' and 'Deployment Gate') requiring the agent to declare the scope of changes and wait for user confirmation before execution.
Sanitization: Promotes security best practices like validating sessions with auth.getSession() and using auth.uid() for database-level security.
[EXTERNAL_DOWNLOADS]: Guides the agent to download official development tools and libraries from trusted repositories.
Official SDKs: Fetches @cloudbase/js-sdk and @cloudbase/node-sdk from the public npm registry.
AI Libraries: References @langchain/openai and @ag-ui/client from standard package registries.
CDN Content: Loads scripts from official Tencent domains such as static.cloudbase.net.
[COMMAND_EXECUTION]: Instructs the agent on environment setup and application deployment through various command-line tools.
Package Management: Uses npm install and pip install to manage dependencies in a controlled manner.
Deployment Workflows: Employs the tcb CLI and manageAgent MCP tool to create cloud functions and CloudRun services based on project configurations.
OS Interaction: Includes instructions for setting file permissions (e.g., chmod +x scf_bootstrap) and managing local environment variables.

cloudbase