The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses forceful imperative language such as 'YOU DO NOT HAVE A CHOICE' and 'not negotiable' to ensure the agent checks for relevant skills before responding. While this employs 'system-level' override patterns, it is mitigated by an explicit priority system that places 'User's explicit instructions' at the highest priority.
[COMMAND_EXECUTION]: Includes read-only shell command examples (e.g., 'git rev-parse', 'git branch --show-current') intended for environment detection in development workflows. These are benign and used to adapt the skill's behavior to the local environment.
[DATA_EXFILTRATION]: Documents tool mappings for network-capable tools like 'WebFetch' and 'WebSearch' to assist in cross-platform portability. No logic for unauthorized data transmission or exfiltration is present.
[SAFE]: The repository provides documentation and tool mappings to maintain consistency across different AI agent environments without introducing malicious code, obfuscated logic, or unauthorized privilege escalation.

using-superpowers