The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses forceful, mandatory language ("ABSOLUTELY MUST", "IF A SKILL APPLIES TO YOUR TASK, YOU DO NOT HAVE A CHOICE", "This is not negotiable") to override default AI behavior and prioritize its own set of rules.
[PROMPT_INJECTION]: Explicitly instructs the agent that "Superpowers skills override default system behavior where they conflict," which is a direct command to bypass standard system prompt constraints.
[COMMAND_EXECUTION]: The platform mapping files (references/codex-tools.md, references/copilot-tools.md, references/gemini-tools.md) provide documentation and direct instructions for using powerful system tools like Bash, run_shell_command, and spawn_agent.
[INDIRECT_PROMPT_INJECTION]: The skill enforces a workflow where the agent is required to load and follow external content (other skills) even if there is only a "1% chance" of relevance, creating a broad surface area for the execution of instructions from potentially untrusted files.
[PERSISTENCE]: In references/codex-tools.md, the skill instructs the user (or the agent) to modify a global configuration file (~/.codex/config.toml) to enable features like multi-agent support, which constitutes a persistent modification of the execution environment.

using-superpowers