The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs extremely forceful and non-negotiable language ("ABSOLUTELY MUST", "not negotiable", "not optional") to override the agent's standard task evaluation and decision-making logic. It specifically provides a "Red Flags" table designed to suppress the agent's internal reasoning (e.g., instructing the agent that thoughts like "I need more context first" or "This is just a simple question" are rationalizations that should be ignored in favor of mandatory tool invocation.
[INDIRECT_PROMPT_INJECTION]: This skill creates a significant vulnerability surface by mandating a "forced-loading" behavior. By commanding the agent to invoke the Skill tool if there is even a "1% chance" a skill might apply, it effectively bypasses the agent's ability to filter potentially malicious or irrelevant content found in the environment.
Ingestion points: The skill mandates the use of the Skill tool to load content from external SKILL.md files or other skill definitions into the active context.
Boundary markers: No boundary markers or instructions are provided to help the agent distinguish between its core system instructions and the potentially untrusted content loaded from the invoked skills.
Capability inventory: While this specific skill only uses the Skill tool, the instructions facilitate the execution of any capabilities possessed by the skills it forces the agent to load (e.g., shell execution, file writes, or network access).
Sanitization: There is no evidence of sanitization, validation, or integrity checking for the skills being invoked; the instruction is to "follow it directly."

using-superpowers