The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess.run function in multiple scripts (backfill_email.py, backfill_calendar.py, backfill_messages.py) to execute local CLI commands. These commands are used to interface with external utilities (gog, wacli) to retrieve historical user data for analysis. The usage does not utilize shell=True and handles arguments safely.
[EXTERNAL_DOWNLOADS]: The skill relies on external command-line utilities that are not included in the standard environment, specifically gog for Google Workspace interaction and wacli for WhatsApp/messaging interaction. While the skill does not download these during runtime, its core functionality depends on their presence.
[PROMPT_INJECTION]: The skill processes untrusted external data from emails, calendar events, and messages to build its causal models, creating an indirect prompt injection surface (Category 8).
Ingestion points: Untrusted data is ingested via scripts/backfill_email.py (Gmail), scripts/backfill_calendar.py (Calendar), and scripts/backfill_messages.py (WhatsApp/Slack/Discord).
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands when parsing the contents of messages or emails.
Capability inventory: The skill possesses shell execution capabilities (via subprocess.run) and file system write access to update its action logs.
Sanitization: No content sanitization or filtering of the retrieved communications is implemented before the data is recorded in the causal log.

causal-inference