The Agent Skills Directory

[COMMAND_EXECUTION]: Path traversal vulnerability in scripts/record_trade.py. The script accepts a user-provided --ts-code argument and uses it to construct a filesystem path without validation or sanitization: md_path = os.path.join(base, "records", f"{args.ts_code}.md"). A malicious input (e.g., using directory traversal sequences) could allow the agent to write or append Markdown content to unauthorized locations on the host system.
[PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill ingests untrusted user data through fields such as 'reason' and 'note' and persists this data in Markdown files and a SQLite database. Because these logs are intended to be read back by the agent for trade review or statistics, malicious instructions embedded in the entries could influence future agent behavior. \n
Ingestion points: Data enters via the --reason, --note, and --ts-code parameters in scripts/record_trade.py. \n
Boundary markers: Absent; the data is appended directly to records without the use of delimiters or instructions to the agent to treat the content as data only. \n
Capability inventory: The skill has the capability to write and create files/directories via scripts/record_trade.py and query a database via scripts/query_trades.py. \n
Sanitization: Absent; the scripts do not perform any escaping, filtering, or validation of user-supplied strings before storage.

stock-trade-journal