google-sheet
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill utilizes the official
googleapisNode.js package, which is a verified and trusted library from Google for interacting with their services. This is the standard and most secure method for such integrations. - [SAFE]: Credential management is implemented following security best practices. The skill avoids hardcoding secrets and instead implements a discovery mechanism that checks environment variables (
GOOGLE_SERVICE_ACCOUNT_KEY) and standard local configuration paths (~/.config/google-sheets/credentials.json), ensuring sensitive keys are handled outside the skill source. - [SAFE]: All network operations are directed to official Google API endpoints through an authenticated client. There is no evidence of unauthorized data exfiltration, suspicious background network calls, or communication with unknown domains.
- [SAFE]: Command implementation in
scripts/sheets.jsinvolves mapping CLI arguments to specific Google Sheets API methods. Input arguments for data and formatting are processed using standard JSON parsing, avoiding dangerous dynamic execution patterns such aseval()or unvalidated command-line construction. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface inherent to tools that ingest external data. This is documented here as a design characteristic for security awareness.
- Ingestion points: Data enters the agent context from external spreadsheets via the
readRangefunction inscripts/sheets.jsusing thespreadsheets.values.getAPI. - Boundary markers: The script does not apply specific delimiters or warnings to the retrieved data to distinguish it from system instructions.
- Capability inventory: The skill has significant capabilities to modify the environment, including writing data, formatting cells, adding/deleting sheets, and resizing layout elements via multiple functions in
scripts/sheets.js. - Sanitization: Retrieved spreadsheet content is returned as raw JSON values without filtering or escaping for potential instructional patterns.
Audit Metadata