The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/registry.py invokes the gcloud CLI tool to obtain transient authentication tokens.\n
Evidence: The adc_token function uses subprocess.run(["gcloud", "auth", "application-default", "print-access-token"], ...) to fetch the ADC token.\n
Context: This is a secure and standard practice for GCP authentication in development environments, ensuring that credentials are not hardcoded or stored insecurely within the skill.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to interact with Google Cloud's Agent Registry service.\n
Evidence: The script makes REST API calls to https://agentregistry.googleapis.com/v1alpha for managing skills and revisions.\n
Context: These requests target a well-known service from a trusted organization (Google) and are necessary for the skill's primary function.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from the GCP Agent Registry, which represents a potential surface for indirect prompt injection from externally stored skill content.\n
Ingestion points: The skill retrieves skill metadata (JSON) and revision content (ZIP) from the agentregistry.googleapis.com API through the list, get, and search commands.\n
Boundary markers: The skill instructions include an explicit "Issue gate" in references/issues.md to guide the agent in handling API inconsistencies, but do not define specific delimiters for separating ingested content from agent instructions.\n
Capability inventory: The skill possesses file system access (read/write for ZIP processing) and network communication capabilities with Google Cloud APIs.\n
Sanitization: Ingested data is handled using standard library functions for JSON and base64; no additional sanitization or filtering of the natural language content within skill descriptions is performed before it is analyzed by the agent.

gcp-agent-registry