The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script scripts/run_deep_research.py uses subprocess.run to execute the core research utility assets/deep_research.py. This implementation is used for internal script delegation and correctly utilizes list-based arguments to prevent shell command injection.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user prompts and incorporates them into a larger instructional context within scripts/run_deep_research.py without using robust boundary markers or sanitization. This creates an attack surface where an adversarial user prompt could attempt to override the intended research parameters or safety constraints of the underlying model.\n
Ingestion points: The prompt positional argument in scripts/run_deep_research.py.\n
Boundary markers: Absent; the script uses simple newline separation for research parameters.\n
Capability inventory: The skill possesses network access to the OpenAI API and file system write access for saving prompts and reports.\n
Sanitization: No input sanitization or filtering is performed on the user-provided prompt before it is passed to the research model.

deep-research