The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user requirements in Phase 1 and use that data to generate technical decompositions, architecture designs, and infrastructure code. This creates a surface for indirect prompt injection where untrusted user input could attempt to influence the generated output. * Ingestion points: User responses to clarifying questions in Phase 1 Step 1 of SKILL.md. * Boundary markers: The instructions do not define explicit boundary markers or delimiters for the user-provided data. * Capability inventory: The skill generates Terraform code in Phase 3 Step 3 and bash/gcloud validation scripts in Phase 4 Step 3. * Sanitization: While the skill recommends using security tools like Model Armor and Cloud DLP in the architecture it designs, it does not specify sanitization steps for its own internal processing of user input.
[DYNAMIC_EXECUTION]: In Phases 3 and 4, the skill generates executable content, including Infrastructure as Code (Terraform) and verification scripts (bash/gcloud), based on the requirements collected earlier. This is the primary function of the skill, and the generated content is intended for user review and manual execution.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and technical samples from several external sources to ground its architectural guidance. * Official Documentation: Retrieves architectural patterns and product guides from docs.cloud.google.com and codelabs.developers.google.com. * Technical Samples: Fetches implementation examples from the official Google samples repository on github.com/google. * Third-Party Documentation: Retrieves configuration guides for database connectivity tools from mcp-toolbox.dev.

google-cloud-solution-agentic-ai-data-science-workflow