The Agent Skills Directory

[Prompt Injection] (LOW): The skill uses aggressive and authoritative markers to bypass the agent's standard operational constraints and logic for tool relevance.
Evidence: Use of phrases like "ABSOLUTELY MUST", "YOU DO NOT HAVE A CHOICE", "not negotiable", and "1% chance" inside <EXTREMELY-IMPORTANT> tags.
Impact: It attempts to prevent the agent from using its own reasoning ("You cannot rationalize your way out of this") to determine if a tool is appropriate, potentially leading to excessive tool usage or vulnerability to further injections if subsequent skills are malicious.
[Indirect Prompt Injection] (LOW): The skill establishes a high-sensitivity trigger where untrusted user input is used to determine which other skills to load and execute.
Ingestion points: User message received flow in the DOT diagram.
Boundary markers: None. The instructions do not provide delimiters or safety warnings for the content of the invoked skills.
Capability inventory: Explicitly requires the use of the Skill tool, which can load and execute additional instructions.
Sanitization: None. The agent is told to "follow it directly" and "Follow skill exactly".

using-superpowers