The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation describes a plugin architecture that uses importlib.import_module(f"broker.{broker}.api.<module>") and getattr(module, "authenticate_broker") to load and execute code based on the configured broker name. This dynamic loading from computed paths is a standard pattern for the skill's plugin-based architecture but requires that the broker variable be strictly validated against a known-safe list.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from various broker APIs, including master contracts (CSV/JSON), order books, and market-data streams. This untrusted data is brought into the agent's context. The analysis found that while schema validation is mentioned for the API layer, there is a lack of explicit instructions for boundary markers or sanitization when this data is presented to the agent, creating a surface where a compromised broker API could influence agent behavior.
[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication material, including BROKER_API_KEY, BROKER_API_SECRET, and encrypted session tokens. While it follows best practices by using environment variables and Fernet encryption for database storage, the skill handles these credentials alongside network capabilities via a shared httpx client, necessitating careful management to prevent accidental exposure.
[COMMAND_EXECUTION]: The documentation contains several instructions involving shell command execution for development and auditing, such as pip download, unzip, and the use of the !fd-audit skill syntax. These commands are intended for environment setup and verification but demonstrate the agent's capability to interact directly with the host system's shell.

broker-integration