watchlist-manager
Warn
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: MEDIUMPERSISTENCECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill includes a dedicated script
scripts/schedule.pydesigned to establish recurring execution across sessions. It programmatically modifies the systemcrontabon Linux and macOS to execute stock monitoring checks every five minutes. On Windows, it provides commands to create persistent tasks viaschtasks. - [COMMAND_EXECUTION]: The skill uses the
subprocessmodule to execute local system utilities. Specifically,scripts/notify.pyinvokes thenotify-sendbinary for desktop alerts, andscripts/schedule.pyexecutes thecrontabcommand to manage the user's scheduled background jobs. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch market data from several established services. It retrieves quotes and earnings information from Yahoo Finance, Alpaca, Finnhub, Alpha Vantage, and Nasdaq. Additionally,
scripts/notify.pytransmits alert data to external endpoints including Slack, ntfy.sh, and user-defined webhooks. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection because it ingests and processes external market data that is summarized for the agent.
- Ingestion points: Financial quotes and earnings reports are fetched from multiple external APIs in
scripts/quotes.py. - Boundary markers: None observed; data is interpolated directly into briefs and alert messages without specific delimiters or instructions to ignore embedded content.
- Capability inventory: The skill can write files (
watchlist.json,.env), perform external network requests (webhooks/APIs), and execute local system commands (crontab,notify-send). - Sanitization: Numerical values are coerced to floats, but textual information from external analysts or providers is not sanitized for potential natural language instructions before being processed.
Audit Metadata