The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to maintain a specific personality and behavioral stance, such as remaining 'awesome regardless of negative input'. Safety is reinforced by a 'Prompt Injection Defense' section in security-protocols.md which mandates treating external content as read-only and explicitly rejects common jailbreak-style patterns like 'Ignore previous instructions'.\n- [DATA_EXFILTRATION]: To protect personal data stored in files like contacts.md, the skill enforces a 'Two Repository Strategy' that isolates sensitive information in a private directory. It also prescribes a pre-commit security checklist to prevent the accidental leakage of secrets using tools like rg.\n- [EXTERNAL_DOWNLOADS]: TOOLS.md provides installation commands for established development runtimes including Bun, uv (for Python), and Rustup. These downloads originate from well-known official domains such as bun.sh, astral.sh, and rustup.rs, which are recognized as trusted technology providers.\n- [COMMAND_EXECUTION]: The infrastructure follows a 'CLI-First' philosophy, which promotes the use of deterministic code over probabilistic AI outputs. This design choice is documented as a means to increase system reliability and ensure that all operations are scriptable, testable, and maintainable.\n- [REMOTE_CODE_EXECUTION]: The system orchestrates specialized agents and CLI tools to perform complex tasks, including security testing via a 'pentester' agent. These interactions are managed through structured workflows and an escalation protocol that includes error detection and recovery mechanisms like checkpoints.

CORE