The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documents hardcoded credentials in SKILL.md, including an Alpaca API Key ID (PK5ZMJM5OF7DAA6EU2PKVXI5Z7) and an Alpaca Secret Key (8mKU43vRGzgM3QFLwdTex1n3Li5FgEGehyVxJJi5Ags1).
[PERSISTENCE]: Automated execution is configured via three specific cron jobs (ef6e544d7bf9, c42aec080bc4, 2f06fc93acda) designed to manage portfolio reporting, weekly reviews, and automated position building.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for process management, specifically running a Flask web server in the background (python3 scripts/portfolio_h5.py &) and terminating it using process signaling (pkill -f portfolio_h5.py).
[TIME_DELAYED_CONDITIONAL]: The skill uses time-based triggers for sensitive operations, such as a scheduled one-time automated trading script (alpaca_build_positions.py) set to execute at a specific future market open date.
[PROMPT_INJECTION]: The skill includes instructions to bypass human-in-the-loop confirmation for trading actions when specific trigger phrases like '你自己定' are used, increasing the agent's autonomy over financial transactions.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection through the ingestion of external data for report generation and dashboard display.
Ingestion points: Fetches market data from Alpaca (paper-api.alpaca.markets), Tencent Finance (qt.gtimg.cn), and exchange rate data (open.er-api.com).
Boundary markers: No explicit instructions or delimiters are provided to the agent to treat data from these external APIs as untrusted or to ignore embedded instructions.
Capability inventory: The agent has capabilities for automated trading via the Alpaca API, web server hosting via Flask, and process management through shell commands.
Sanitization: There is no evidence of sanitization or validation of the data received from external APIs before it is incorporated into reports or displayed on the H5 dashboard.

us-stock-trading