gemini-files-api

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the macOS security command to fetch API keys from the system keychain if they are not provided via environment variables. This is a local operation for secure credential management.\n- [EXTERNAL_DOWNLOADS]: The scripts/bootstrap.sh script installs the @google/genai library from the npm registry using npm ci --ignore-scripts, which prevents the execution of potentially malicious scripts packaged with dependencies.\n- [DATA_EXFILTRATION]: The skill transmits local files to Google's Gemini API for processing. This behavior is the primary intended function of the tool and targets a well-known service provider.\n- [PROMPT_INJECTION]: The skill provides an interface for processing untrusted multimodal data which represents an indirect prompt injection surface.\n
  • Ingestion points: scripts/gemini-mm.mjs accepts local file paths via the --file argument.\n
  • Boundary markers: The skill uses the official Google SDK's createUserContent and createPartFromUri functions to structure the request, which helps separate the user prompt from the file content.\n
  • Capability inventory: Network access to Google Gemini API endpoints.\n
  • Sanitization: No content filtering is performed by the skill; it relies on the downstream LLM's safety filters to manage malicious content within processed files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 04:11 PM
Security Audit — agent-trust-hub — gemini-files-api