The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses directive language to override default agent planning behavior, requiring the agent to invoke tools for any task with a 1% relevance and explicitly forbidding the agent from opting out of this workflow. This is mitigated by the hierarchical instruction structure that places user-provided commands at the highest priority.
[SAFE]: The skill establishes an indirect prompt injection surface by requiring the agent to load and process external instructions (skill files) in response to user messages. Ingestion points include user task descriptions, and the capability inventory includes platform-specific tools for file access and subagent creation. The risk is managed through a priority-based boundary system and reliance on platform-native skill loading.
[SAFE]: The skill includes reference documentation that maps generic agent actions to specific CLI tools for platforms like Claude Code, Gemini CLI, and Copilot CLI, facilitating standard developer operations without introducing malicious behavior.

using-superpowers