The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses extremely forceful directives ("EXTREMELY-IMPORTANT", "ABSOLUTELY MUST", "non-negotiable") to mandate agent behavior and explicitly states that "Superpowers skills override default system behavior." It requires the agent to invoke the Skill tool before any other action, including responding to the user or asking clarifying questions.
[PROMPT_INJECTION]: The skill mandates a "1% chance" threshold for invoking external skills, which creates a surface for indirect prompt injection by encouraging the loading of instruction sets with minimal relevance checking. 1. Ingestion points: External skill files loaded via the Skill or activate_skill tools. 2. Boundary markers: Absent; the agent is instructed to "Follow skill exactly" without specific warnings about untrusted content. 3. Capability inventory: The skill maps capabilities like file writing, shell access, and subagent creation. 4. Sanitization: Absent.
[COMMAND_EXECUTION]: The reference files map powerful platform tools such as shell access (Bash/run_shell_command) and subagent management (Task/spawn_agent), which increase the potential impact if the skill loading mechanism is exploited.

using-superpowers