The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to download the gogcli binary from a personal GitHub repository (github.com/steipete/gogcli). This source is not categorized as a trusted organization or a recognized well-known service in the security baseline.\n- [COMMAND_EXECUTION]: The skill requires the execution of external binaries (gogcli), shell scripts (build-safe.sh), and potentially destructive file system operations such as Remove-Item -Recurse -Force used during cleanup and recovery drills.\n- [PERSISTENCE]: The skill establishes persistence by configuring automated backup routines through the Windows Task Scheduler and cron utility to ensure daily and weekly operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources including Gmail and Google Drive, creating a surface for indirect prompt injection if an agent subsequently processes or interprets the content of the decrypted backup files.\n
Ingestion points: Data is ingested from Google Workspace services (Gmail, Calendar, Drive) via the gog backup push and gog backup cat commands.\n
Boundary markers: The instructions do not define explicit delimiters or security markers to ensure the agent ignores embedded instructions within the backed-up data.\n
Capability inventory: The skill is granted Bash, Read, and Write tool permissions, allowing for significant system interaction based on data inputs.\n
Sanitization: There is no description of content filtering or sanitization for the data retrieved from external Google services.

matematic-workspace-backup