The Agent Skills Directory

[Command Execution] (LOW): The test suite (tests/test_cli.py, tests/test_bank_cli.py, etc.) uses subprocess.run to execute the CLI script during verification. This is standard behavior for testing CLI tools and is restricted to the skill's own scripts.
[Dynamic Execution] (LOW): The scripts/akshare_cli.py script uses getattr() to dynamically map user-provided function names to the akshare library. This is safely scoped to the ak module namespace and incorporates type-checking via inspect.signature to ensure arguments match expected Python types.
[Indirect Prompt Injection] (LOW): The skill processes data from external financial sources and NLP APIs (e.g., nlp_answer, nlp_ownthink).
Ingestion points: External data enters the system from various financial web portals and the OwnThink knowledge graph service.
Boundary markers: No explicit LLM boundary markers are used, but the CLI provides structured output (JSON, CSV, or formatted tables) which facilitates safe parsing by an agent.
Capability inventory: The skill is a data reader; it does not possess file-writing or arbitrary system execution capabilities.
Sanitization: Data is processed and formatted using pandas and json.dumps, providing consistent structure and reducing the risk of injection during rendering.
[External Downloads] (SAFE): The skill connects to various legitimate financial data providers (Sina, CBIRC, Jin10, etc.) as intended for its primary purpose.

akshare-skill