The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation includes examples using the Bash tool to create configuration directories (.claude/agents/) and write markdown files that define sub-agent behaviors. These commands are transparent, local to the project environment, and necessary for the stated purpose of agent management.
[PROMPT_INJECTION]: Analysis of the example system prompts for sub-agents (such as 'reviewer' and 'debugger') shows they are designed for legitimate engineering tasks. There are no attempts to subvert agent instructions, extract sensitive system prompts, or bypass safety guidelines.
[DATA_EXFILTRATION]: No network operations to unknown external domains or attempts to access sensitive system files (like SSH keys or credentials) were detected. The file operations are restricted to the local project configuration directory.

sub-agents