The Agent Skills Directory

[SAFE]: The skill is purely instructional and serves as a management layer for the Superpowers framework. It documents how to interact with the skill system across different runtimes without performing unauthorized actions.
[PROMPT_INJECTION]: While the skill uses strong directives to enforce its workflow (e.g., "not negotiable," "not optional"), it includes a safety-oriented priority model where user-supplied instructions (CLAUDE.md, direct requests) always override the skill's own logic.
[SAFE]: References to standard application paths (e.g., ~/.agents/skills/ or ~/.claude/skills/) are used for platform awareness and do not constitute exposure of sensitive credentials or private user data.
[PROMPT_INJECTION]: The skill facilitates an architectural surface for indirect prompt injection by design, as its primary purpose is to load and execute instructions from external skill files.
Ingestion points: SKILL.md files located in project-level or global skill directories.
Boundary markers: Not explicitly defined in this skill; the framework relies on platform-level tool invocation (e.g., activate_skill).
Capability inventory: Includes full platform capabilities such as shell command execution, file system modification, and web fetching as documented in the provided reference mappings.
Sanitization: No sanitization of skill content is described within these logical instructions.
[COMMAND_EXECUTION]: The skill documents platform-specific shell tools (e.g., Bash, run_shell_command, shell) for reference purposes. It does not contain any hidden or automated shell commands that execute without user context or visibility.

using-superpowers