The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires installing a third-party tool gogcli from a personal Homebrew tap (steipete/tap/gogcli). This introduces an external dependency from a third-party repository.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from external Google Workspace sources, which could contain malicious instructions.
Ingestion points: Data is retrieved from Gmail, Drive, Docs, and Sheets using gogcli commands defined in SKILL.md.
Boundary markers: The instructions do not define clear delimiters or instruct the agent to ignore instructions embedded within the retrieved data.
Capability inventory: The skill has access to Bash and Write tools, enabling it to execute commands and write files based on potentially manipulated model output.
Sanitization: No specific sanitization or filtering of retrieved content is mentioned, although the skill does include a mitigation requiring user confirmation before sending emails.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute gogcli commands. This provides the agent with shell access, which could be abused if the agent's context is compromised by malicious instructions in data or prompts.

google-workspace