The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill operates by dynamically generating Python code based on user requests and then executing that code using the Bash tool. This runtime code generation and execution is a core feature but inherently carries risk as the execution flow is determined at runtime.
[COMMAND_EXECUTION]: The skill requests and utilizes broad Bash(cmd:*) permissions. This level of access allows for arbitrary command execution on the host system, which is used here to run the generated data processing scripts.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its data processing nature.
Ingestion points: The skill reads untrusted external data from files like 'my_data.csv' and 'large_dataset.csv' as described in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters defined to separate data content from instructions during processing.
Capability inventory: The skill has high-privilege capabilities including Bash(cmd:*), Read, Write, and Edit permissions, and it explicitly generates and runs code.
Sanitization: There is no evidence of sanitization or validation logic to prevent malicious instructions embedded within datasets from influencing the agent's behavior.

splitting-datasets