docs
Warn
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The instructions in SKILL.md and references/commands.md direct the agent to execute code using 'cargo run --' or a locally built debug binary. This process involves runtime compilation and execution of source code contained within the repository environment.\n- [COMMAND_EXECUTION]: The skill utilizes shell command execution to interface with the goog CLI tool for all operations, including authentication management (goog auth login) and document manipulation.\n- [PROMPT_INJECTION]: The skill processes data from external and local sources that could contain malicious instructions designed to influence the agent's behavior.\n
- Ingestion points: The skill reads local files (e.g., metrics.csv) and retrieves content from Google Docs (goog docs get/map).\n
- Boundary markers: The provided instructions lack explicit delimiters or instructions to disregard directives embedded within the ingested content.\n
- Capability inventory: The agent has the ability to execute shell commands, access the local file system, and perform network operations to interact with Google APIs.\n
- Sanitization: There is no documentation or evidence of input validation, escaping, or filtering of the external content before it is processed or written to documents.
Audit Metadata