The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs strong imperative language and behavioral overrides to bypass the agent's standard task assessment and reasoning. In SKILL.md, it uses "EXTREMELY-IMPORTANT" blocks to mandate skill calls even for cases where the agent sees only a "1% possibility" of applicability, stating this is "non-negotiable". Furthermore, the "Red Lines" section explicitly instructs the agent to disregard its own rationalizations (e.g., "this is just a simple question") and instead strictly follow the skill's rules. It also uses a <SUBAGENT-STOP> tag to instruct sub-agents to "ignore this skill".\n- [COMMAND_EXECUTION]: The reference files contain templates for executing system-level commands across multiple CLI platforms. For instance, references/copilot-tools.md provides instructions for terminating processes on Windows using Stop-Process -Id <PID> and execution patterns for running shell scripts via PowerShell. Additionally, references/codex-tools.md includes a shell script snippet for Git environment detection.

using-superpowers