The Agent Skills Directory

[COMMAND_EXECUTION]: The rules/assets/core/portfolio_runner.py script uses subprocess.Popen to manage and execute child strategy processes. The specific scripts and their environment variables are determined by a YAML configuration file, which allows for the execution of arbitrary Python scripts within the environment.
[REMOTE_CODE_EXECUTION]: The machine learning strategy templates (rules/assets/ml_logistic/strategy.py and rules/assets/ml_xgb/strategy.py) utilize the pickle module to load pre-trained model pipelines. Using pickle.load() on a file is a known security risk as it can be used to execute arbitrary code if the pickle file is maliciously crafted.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface that could be used for indirect injection if market data sources are compromised.
Ingestion points: Market data is fetched via client.history() (OpenAlgo API) and through direct DuckDB file connections in rules/assets/core/data_router.py.
Boundary markers: No explicit boundary markers or 'ignore' instructions for data streams were observed.
Capability inventory: The skill can execute child processes via subprocess.Popen in portfolio_runner.py, perform file writes for state management in state.py, and initiate network requests via the OpenAlgo SDK.
Sanitization: Market data and database records are processed as numerical and categorical data without explicit sanitization or validation of the content before it influences strategy execution logic.

algo-expert