The Agent Skills Directory

[SAFE]: The skill implements a strong security policy for subagent delegation, including explicit 'least privilege' constraints and a strict prohibition on recursive delegation.
[SAFE]: Explicit data protection measures are included, specifically instructing the parent agent to never share secrets, tokens, or environment files with children.
[SAFE]: The instructions include a mandatory cleanup protocol that requires stopping subagent processes and removing temporary resources like sockets or worktrees after task completion.
[SAFE]: A 'Verify + integrate' phase is defined which requires the parent agent to inspect diffs and rerun validation checks locally rather than trusting subagent claims blindly.
[SAFE]: Subagent roles (Scout, Research, Worker) are specialized and restricted; for example, 'Worker' agents are limited to an authorized directory and forbidden from mutating version control systems.

use-subagents