The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes dynamic module discovery and loading in scripts/views/registry.py. It uses importlib.import_module and pkgutil.walk_packages to import all Python modules within the scripts/views/ directory at runtime to build its view registry.
[DYNAMIC_EXECUTION]: In scripts/views/registry.py, the function _build_tool_view_module creates new ModuleType objects dynamically to wrap AKShare tool definitions into the toolkit's view architecture.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external financial platforms, specifically investor comments and sentiment rankings in scripts/views/hot_rank_sentiment_dashboard.py (via tools like stock_comment_detail_scrd_focus_em).
Ingestion points: External data is ingested through the akshare library from third-party financial websites.
Boundary markers: None. The raw text data is returned to the agent within a JSON envelope.
Capability inventory: The skill has network access (remote API mode), file system write access (caching mechanism), and can execute Python logic.
Sanitization: There is no evidence of text sanitization or filtering to prevent the agent from following instructions embedded in the financial comments or descriptions.
[DATA_EXFILTRATION]: The scripts/views_runner.py script enables a remote operation mode where it connects to a URL specified by the FINSKILLS_VIEW_API_URL environment variable. It is designed to transmit sensitive authentication tokens (FINSKILLS_VIEW_API_TOKEN or FINSKILLS_VIEW_API_KEY) to these arbitrary remote endpoints.
[COMMAND_EXECUTION]: The scripts/akshare_tools.py script implements a dynamic dispatch mechanism that allows an agent to call any function within the akshare library by name, provided it is listed in the local config/litellm_tools.json registry.

findata-toolkit-cn