The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of a third-party global npm package @mariozechner/gdcli from the public npm registry, which is not a verified vendor repository.
[CREDENTIALS_UNSAFE]: The skill instructions identify and interact with sensitive local file paths used for storing OAuth client secrets and access tokens: ~/.gdcli/credentials.json and ~/.gdcli/accounts.json.
[COMMAND_EXECUTION]: The skill executes shell commands using the gdcli tool, incorporating user-provided parameters for email addresses, file IDs, and search queries, which could be exploited for command injection if input is not properly handled by the underlying tool.
[DATA_EXFILTRATION]: The skill includes functionality to share Google Drive files publicly using the share <fileId> --anyone command, which represents a risk of accidental data exposure if the agent is manipulated into sharing sensitive files.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via the following attack surface:
Ingestion points: The <query> and <text> parameters used in the ls --query and search commands.
Boundary markers: No delimiters or instructions are provided to the agent to treat the user-supplied query content as untrusted data.
Capability inventory: The skill possesses the capability to list, search, download, and publicly share files from the user's Google Drive.
Sanitization: There is no evidence of sanitization or validation performed on the query strings before they are passed to the CLI tool.

gdcli