The Agent Skills Directory

Local Configuration Modification: The skill instructs the agent to locate and modify sensitive local configuration files such as ~/.gemini/config/mcp_config.json, ~/.codeium/windsurf/mcp_config.json, and claude_desktop_config.json. This pattern is used to automatically add the Google Cloud Monitoring MCP server, but it involves the agent writing to files that control its own execution environment and tool availability.
Environment Context Access: Instructions direct the agent to search for GCP Project IDs within the "environment context." This involves reading existing environment variables or context data, which is a standard requirement for cloud authentication but constitutes a broad data access pattern.
External Tool Integration: The skill configures and interacts with a remote service at https://monitoring.googleapis.com/mcp. This is a well-known service provided by the vendor (Google), and the integration is used to fetch live metric descriptors. Per standard security evaluation, this connection to an official vendor endpoint is considered a functional component of the skill.
Indirect Prompt Injection Surface: The skill processes user prompts to extract keywords and resource identifiers which are then interpolated into API filters. While common for search-based skills, this creates a data ingestion surface where external input influences tool parameters. The skill currently lacks explicit boundary markers or sanitization instructions for this interpolated data.

cloud-monitoring-metric-selection