The Agent Skills Directory

[SAFE]: The skill functions as a local configuration auditor and linter for AI agent definitions. It is designed to detect and remediate security risks such as the use of bypassPermissions or unconstrained tool access.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute find commands for discovering agent configuration files within the project's .claude/agents/ directory and the user's ~/.claude/agents/ directory. These commands are static and limited to discovery.
[DATA_EXPOSURE]: The skill accesses agent metadata and system prompts stored in local markdown files. This data is processed locally to generate scores and recommendations for the user. No network tools are requested in the allowed-tools section, and no data exfiltration patterns were identified.
[REMOTE_CODE_EXECUTION]: There are no patterns of downloading external scripts or executing remote code. All operations are performed on existing local files.

eval-agents