The Agent Skills Directory

[SAFE]: The skill provides a controlled environment for running subagents in parallel, incorporating several security best practices such as mandatory safety instructions for child prompts and task isolation using temporary directories. It explicitly prohibits subagents from interacting with external systems (e.g., posting, messaging, or mutating issues) and restricts their communication to other agents only.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns or unsafe external fetches were detected. The skill references an external repository for attribution in its notices, but does not perform any runtime downloads or execution of external scripts.
[DATA_EXFILTRATION]: The skill includes explicit instructions to keep all findings and drafts local and forbids any external communication or messaging, mitigating the risk of data exfiltration.
[PROMPT_INJECTION]: No malicious prompt injection or override patterns were found. The instructional content is focused on operational reliability and safety constraints for subagents.

swarm