The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources including Google Doc content (scripts/gdrive_transcript_search.py), Gmail snippets (scripts/gmail_search.py), and Calendar event descriptions (scripts/google_calendar_search.py). Malicious instructions embedded in these external sources could influence the agent's behavior, especially given the skill's capabilities to create folders and upload files.
Ingestion points: get_file_content in scripts/gdrive_transcript_search.py reads document text; search_messages in scripts/gmail_search.py reads email snippets.
Boundary markers: None identified; content is passed directly to the agent context.
Capability inventory: File upload (scripts/google_drive_upload.py), folder creation (scripts/create_client_folder.py), and recursive search across Drive, Gmail, and Calendar.
Sanitization: No sanitization or validation of the ingested content is performed before processing.
[DATA_EXPOSURE]: The skill manages highly sensitive data (Gmail, Drive, Calendar) and uses local files to store OAuth credentials and refresh tokens (client_secrets.json, mycreds.txt, gmail_token.pickle, calendar_token.pickle). Compromise of the local environment would grant an attacker persistent access to the user's Google account.
[REMOTE_CODE_EXECUTION]: The scripts scripts/gmail_search.py and scripts/google_calendar_search.py use the pickle module to deserialize locally stored credentials. While these files are generated by the script itself, the use of pickle.load() on the local filesystem represents a potential risk if the environment is tampered with.
[EXTERNAL_DOWNLOADS]: The skill depends on standard, well-known libraries such as pydrive2, google-auth, and google-api-python-client. These are legitimate dependencies for interacting with Google's APIs.

google-workspace