The Agent Skills Directory

[PERSISTENCE]: The skill includes an optional installer (optional-workspace-mcp/install-autostart.sh) that sets up a macOS LaunchAgent or Linux systemd user unit to manage the transcript-fetching server. This persistence mechanism is the primary purpose of the script, is fully documented, requires explicit interactive user consent during installation, and provides a corresponding uninstaller.
[COMMAND_EXECUTION]: Local service management scripts (start.sh, stop.sh, doctor.sh) and the provided test suite (test_doctor.py, test_transcript_primary.py) use shell commands to coordinate process lifecycles and validate component behavior. These operations are confined to the user's local environment and serve legitimate maintenance and testing roles.
[EXTERNAL_DOWNLOADS]: The start.sh helper utilizes the uvx package runner to execute the third-party workspace-mcp utility from a public registry. This dependency is transparently declared as an optional component for users requiring advanced multi-account Google Workspace integration.
[INDIRECT_PROMPT_INJECTION]: Because the skill fetches and stores meeting content from external services (Gmail and Google Drive), it introduces a surface for indirect prompt injection. Malicious instructions embedded within a transcript could potentially influence an agent in subsequent tasks. The skill mitigates this by focusing on structural and provenance verification through its verify_transcripts.py and transcript_primary.py components, though it does not implement content-based sanitization.

synthesis-meeting-transcripts