The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the gog binary from an external GitHub repository (github.com/openclaw/gogcli) which is not on the trusted vendor list.
Evidence: SKILL.md and references/configuration.md provide shell commands to fetch the binary from a GitHub release URL (https://github.com/openclaw/gogcli/releases/download/...).
[REMOTE_CODE_EXECUTION]: The skill's operational model is based on downloading and executing an external binary (gog) from a third-party source. Running unvetted code from the internet poses a security risk to the environment.
Evidence: All core instructions in SKILL.md and referenced files call the gog command, which is an externally sourced binary.
[COMMAND_EXECUTION]: The agent is granted permission to execute a wide variety of commands via the gog CLI, which has extensive access to the user's Google account data and services.
Evidence: The skill configuration in SKILL.md includes allowed-tools: Bash(gog:*), enabling full access to the tool's command set.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content from potentially malicious external sources (like incoming emails or shared documents) while possessing the authority to take actions like exfiltrating data or deleting files.
Ingestion points: The skill reads data using gog gmail search, gog gmail get, gog drive search, gog drive get, gog docs cat, and gog sheets get.
Boundary markers: The instructions lack specific guidance or delimiters to prevent the agent from interpreting embedded instructions within the retrieved content as its own commands.
Capability inventory: The skill includes powerful write capabilities such as gog gmail send, gog gmail forward, gog drive share, gog drive delete, gog drive upload, and gog docs clear.
Sanitization: There are no mentioned mechanisms for sanitizing or escaping the untrusted data before it is processed or used in further commands.

google