The Agent Skills Directory

PROMPT_INJECTION (SAFE): The user input gathered in scripts/run_deep_research.py is intended to be the actual prompt for a research agent. While this is an indirect ingestion surface, it is the primary function of the tool. There are no attempts to bypass safety filters or ignore prior instructions in the skill itself.
DATA_EXPOSURE (SAFE): The skill uses a .env file to manage the OPENAI_API_KEY. This is standard developer practice. No evidence was found of exfiltrating sensitive local files or hardcoding real secrets.
UNVERIFIABLE_DEPENDENCIES (SAFE): The project depends on the official openai Python package. No malicious remote script executions or suspicious third-party packages were detected.
COMMAND_EXECUTION (SAFE): scripts/run_deep_research.py uses subprocess.run to call assets/deep_research.py. This is used for legitimate internal orchestration and uses argument lists to prevent shell injection.
PERSISTENCE (SAFE): No attempts to modify system startup files or cron jobs were detected.
OBFUSCATION (SAFE): The code is clear and does not use encoding or hidden characters to mask behavior.

deep-research