gemini-api-dev
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill uses explicit instructions to override the agent's internal knowledge base, specifically stating 'These rules override your training data' and 'Your knowledge is outdated' to prioritize the skill's content over previous training.
- [EXTERNAL_DOWNLOADS]: The skill fetches documentation indexes and model specification files from official Google domains including 'ai.google.dev'.
- [COMMAND_EXECUTION]: Instructions provided for the installation of development SDKs using package managers such as 'pip', 'npm', 'go', and 'maven'.
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and process external content from an API documentation index, creating a vulnerability surface where the agent could ingest instructions from remote sources.
- Ingestion points: Documentation index located at 'https://ai.google.dev/gemini-api/docs/llms.txt' and associated markdown files.
- Boundary markers: None identified; instructions do not specify delimiters to separate fetched documentation from system instructions.
- Capability inventory: The skill facilitates file creation (code examples), network access (documentation fetching), and package installation commands.
- Sanitization: No sanitization or validation protocols are defined for the content retrieved from the external documentation index.
Audit Metadata