The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides tools and instructions for reading content from Google Cloud Storage objects (e.g., using read_object or gcloud storage cat). This ingested data could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: External data is brought into the context via object read operations described in references/cli-api-usage.md and references/mcp-usage.md.
Boundary markers: The prompt lacks specific instructions or delimiters to isolate untrusted object content from the agent's primary instructions.
Capability inventory: The skill allows the agent to execute shell commands (gcloud), perform network requests (curl), and write to the local file system (download_object).
Sanitization: No sanitization or verification steps are specified for the data retrieved from external objects before it is processed by the agent context.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download an executable binary from https://storage.googleapis.com/mcp-toolbox-for-databases/. This source is an official Google Cloud domain and is considered trusted.
[EXTERNAL_DOWNLOADS]: The skill includes commands to install related extensions from the vendor's repository using npx skills add gemini-cli-extensions/google-cloud-storage. These resources are provided by the skill's verified author.
[COMMAND_EXECUTION]: The skill relies on shell command execution for its core functionality, including the use of gcloud, curl, and npx for resource management and environment setup.

google-cloud-storage-basics