The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill encourages the installation of arbitrary Python packages via pip install when it encounters an ImportError. These packages are determined by the agent from runtime error logs rather than a fixed, verified manifest, introducing a risk of malicious package installation.- [COMMAND_EXECUTION]: The skill uses bash() to execute shell commands for package installation and code validation tasks.- [CREDENTIALS_UNSAFE]: The skill accesses config.json specifically to diagnose issues related to API tokens, which confirms it interacts with files likely containing sensitive authentication credentials.- [DYNAMIC_EXECUTION]: The skill executes dynamically constructed Python code using a shell command to validate the abstract syntax tree (AST) of user-modified files.- [INDIRECT_PROMPT_INJECTION]: The skill parses untrusted local files including code, configuration, and data artifacts to drive its automated repair logic, creating an attack surface where malicious input could manipulate the agent's actions.
Ingestion points: artifacts/metrics.csv, equity.csv, trades.csv, code/signal_engine.py, and config.json.
Boundary markers: None. No delimiters are used to separate untrusted file content from diagnostic instructions.
Capability inventory: edit_file, bash(), and refresh_strategy_evidence.
Sanitization: None. The skill processes the raw contents of files to identify and fix bugs.

backtest-diagnose