The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes multiple Bash and Python scripts to interface with external AI binaries like Claude Code and Codex. Security analysis of the Python wrappers confirms that subprocesses are called using list-based arguments rather than shell strings, which is a robust defense against command injection vulnerabilities during orchestration.\n- [DATA_EXFILTRATION]: The gpt-pro-bundle.sh script allows for the packaging of local codebase context for analysis by external deep-reasoning models. This workflow is protected by a built-in security scanner (SECRET_RE) that identifies private keys, AWS access keys, and various API tokens, blocking the export process if sensitive data is detected.\n- [EXTERNAL_DOWNLOADS]: The skill references several third-party tools and libraries, including the unofficial notebooklm-py. The documentation provides explicit warnings regarding the unofficial nature of these dependencies and their use of undocumented APIs, ensuring the user is informed of the associated stability and privacy risks.\n- [PRIVILEGE_ESCALATION]: To support autonomous agent-to-agent operations, the skill employs CLI flags such as --dangerously-bypass-approvals-and-sandbox for Codex and --trust-all-tools for Kiro. While these flags bypass interactive confirmation steps, their use is limited to the delegated agent's context and is a necessary trade-off for the skill's primary automation use-case.

call-agent