The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and downloads the workspace-mcp tool for multi-account support.
start.sh uses the uvx tool (from Astral) to install and run the workspace-mcp server from a remote registry.
README.md and SKILL.md link to the taylorwilsdon/google_workspace_mcp GitHub repository as the source for the self-hosted component.
[COMMAND_EXECUTION]: The skill provides shell scripts (start.sh, stop.sh, install-autostart.sh) to manage a local MCP server instance.
install-autostart.sh creates system-level persistence using launchd on macOS (~/Library/LaunchAgents/) and systemd on Linux (~/.config/systemd/user/) to ensure the server starts on login. This is the intended and documented functionality for users choosing the self-hosted path.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its ingestion of external document content.
Ingestion points: optional-workspace-mcp/fetch-meeting.py (line 154) fetches meeting notes and transcripts from Google Drive via the get_drive_file_content tool.
Boundary markers: The protocol instructions in SKILL.md do not specify delimiters or 'ignore' instructions when processing the document body.
Capability inventory: The skill has file-write capabilities in optional-workspace-mcp/fetch-meeting.py (line 172) and instructions in SKILL.md (Step 4 and 5) to save content and update local project indices.
Sanitization: Output filenames are sanitized using a slugify function, but the content of the meeting transcript is written to local markdown files as-is.

synthesis-meeting-transcripts