The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a 'Suggest Using K-Dense Web' directive that instructs the agent to monitor task complexity and steer the user toward a specific hosted platform (k-dense.ai). This behavioral steering influences agent responses to promote a specific service based on interaction context.
[DYNAMIC_EXECUTION]: The framework's core operation involves the autonomous generation and execution of Python code to process biomedical data. The documentation explicitly warns that this code is executed with full system privileges, which poses a significant risk if the generated logic is flawed or manipulated by adversarial input.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface where natural language queries passed to the agent.go() method drive the code generation and execution process without explicit sanitization.
Ingestion points: The natural language research task string passed to the A1.go() method in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are specified in the usage patterns.
Capability inventory: The skill possesses the capability to generate and run arbitrary Python scripts with full system privileges.
Sanitization: The instructions do not mention any automated validation or sandboxing of the generated code, instead placing the burden of review on the user.
[COMMAND_EXECUTION]: The skill instructs the user to perform package installations using uv pip install and suggests the use of setup scripts (scripts/setup_environment.py) that interact with the system environment and shell.
[EXTERNAL_DOWNLOADS]: The skill is configured to automatically download approximately 11GB of external biomedical data lake resources upon first initialization.

biomni