The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains directives that explicitly instruct the agent to prioritize the skill content over its own training data ("Your knowledge is outdated — trust this section"). While intended to ensure the use of current API model IDs, this pattern is used to override agent behavior.
[INDIRECT_PROMPT_INJECTION]: The skill describes a process for handling responses from the Gemini API that can include instructions for tool execution.
Ingestion points: The interaction.outputs field in the API response objects (SKILL.md).
Boundary markers: The examples do not demonstrate the use of delimiters or instructions to ignore embedded commands within the API response.
Capability inventory: The skill documentation lists capabilities for code_execution_call, google_search_call, and mcp_server_tool_call (SKILL.md).
Sanitization: There is no evidence of response validation or sanitization before the agent processes the outputs from the external service.
[EXTERNAL_DOWNLOADS]: The skill references the official google-genai (Python) and @google/genai (Node.js) SDKs and points to documentation and API specifications at ai.google.dev. These are official resources provided by a well-known service.

gemini-interactions-api