The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill implements execute_terminal_command and execute_run_configuration tools. These allow the agent to run any command or script within the IDE's terminal environment. While this is the primary purpose of the skill, it represents a high-risk capability. The severity is set to MEDIUM because the documentation explicitly warns about the risks and the 'Brave Mode' which disables safety confirmations.
DATA_EXFILTRATION (MEDIUM): Tools such as get_file_text_by_path and search_in_files_by_text allow the agent to read project files. This could lead to the exposure of sensitive data like .env files, API keys, or internal documentation if the agent is misdirected.
PROMPT_INJECTION (LOW): The instructions describe 'Brave Mode' and how to enable/disable it. An attacker could potentially use indirect prompt injection in a project file to trick the agent into enabling this mode or executing destructive commands under the guise of automation.
Indirect Prompt Injection (LOW):
Ingestion points: The agent reads untrusted data via get_file_text_by_path, search_in_files_by_text, and get_symbol_info.
Boundary markers: None are specified in the provided instruction files.
Capability inventory: The skill possesses extensive capabilities including execute_terminal_command, create_new_file, and replace_text_in_file.
Sanitization: No sanitization or validation of the content read from files is mentioned before the agent processes it.

jetbrains-skill