The Agent Skills Directory

[SAFE]: The skill provides documentation and templates for configuring subagents within the standard directories used by the Cursor editor (.cursor/agents/ and ~/.cursor/agents/).
[COMMAND_EXECUTION]: The skill uses standard, non-malicious shell commands such as mkdir, touch, and git diff to facilitate project configuration and task analysis.
[INDIRECT_PROMPT_INJECTION]: The skill mentions inferring subagent behavior from conversation context. While this is an ingestion point for untrusted data, it is a core functional requirement for the skill's purpose and does not include malicious override patterns.
[DATA_EXPOSURE]: The 'Data Scientist' example includes usage of the BigQuery CLI (bq), which is a powerful tool for data analysis but is used here in a manner consistent with the subagent's intended task and does not show signs of exfiltration.

create-subagent