The Agent Skills Directory

[SAFE]: The skill acts exclusively as a documentation and reference guide. It does not contain any executable scripts, remote code, or malicious instructions. All content is educational in nature, focusing on the configuration of platform-native subagents.\n- [COMMAND_EXECUTION]: The documentation describes how to grant agents access to the Bash tool and configure lifecycle hooks (e.g., PreToolUse). It emphasizes security best practices by demonstrating tool allowlisting (e.g., Bash(npm *)) to limit an agent's command-line capabilities.\n- [EXTERNAL_DOWNLOADS]: Includes examples referencing well-known and official tools, such as @playwright/mcp for UI testing. These are standard development tools and are documented as part of an optional feature set.\n- [DATA_EXFILTRATION]: Documents the persistent memory features of the platform, which utilize local directories (e.g., ~/.claude/agent-memory/) for state storage. The skill does not contain instructions or logic for sending sensitive data to external or untrusted domains.\n- [PROMPT_INJECTION]: While the skill explains how agents process instructions, it does not include patterns aimed at bypassing platform safety filters or overriding system behaviors in a malicious way. It provides standard operational guidance for agent discovery and delegation.

agent-creator