gemini-cli

Warn

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONMETADATA_POISONINGCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill allows the ingestion of untrusted data from local files through the --context flag in the scripts/gemini.ts script. This data is passed to the AI model without proper isolation, potentially allowing malicious content within those files to hijack the agent's instructions.
  • Ingestion points: The --context parameter in scripts/gemini.ts enables the reading of external file content into the model's active context.
  • Boundary markers: The script does not implement any delimiters or "ignore" instructions to separate the provided file context from the system instructions, increasing the risk of the model following instructions hidden in the data.
  • Capability inventory: The skill possesses the capability to execute system commands via child_process.spawn and write to the local filesystem using fs.writeFileSync.
  • Sanitization: No sanitization, validation, or escaping of the file content is performed before it is interpolated into the CLI command arguments.
  • [METADATA_POISONING]: The skill's documentation and metadata make claims that the tool is 'Google's official command-line interface' and provide a link to https://geminicli.com/. However, official Google AI tools typically reside on google.com domains, and the skill is authored by a third party. This impersonation represents a deceptive behavior that could lead users to trust the tool with sensitive credentials.
  • [COMMAND_EXECUTION]: The script scripts/gemini.ts uses child_process.spawn to interact with the underlying gemini executable. While it uses an argument array to prevent basic shell injection, it still provides a direct bridge for the agent to execute complex CLI operations on the host system based on processed prompts.
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to install the @google/gemini-cli package from npm. While the package name uses a well-known organization scope, the discrepancy between the claimed official status and the non-standard homepage URL suggests potential for supply chain confusion.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 9, 2026, 03:40 PM