The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses extremely forceful and non-negotiable language ("ABSOLUTELY MUST", "not negotiable", "not optional") to override the agent's default decision-making process. It mandates tool invocation even for simple clarifications, which is a technique that can be used to hijack the agent's standard reasoning loop.
[PROMPT_INJECTION]: The "Red Flags" section explicitly instructs the agent to ignore its own internal logic or "rationalization" (e.g., "This is just a simple question"), effectively instructing the agent to bypass its intended flexible response behavior in favor of the skill's rigid rules.
[PROMPT_INJECTION]: The instruction to "Follow skill exactly" and "Don't adapt away discipline" for rigid skills encourages the agent to prioritize skill instructions over potential safety guardrails or user-provided constraints.
[INDIRECT_PROMPT_INJECTION]: The skill mandates the loading of external content (other skills) based on a very low threshold ("1% chance"), which significantly increases the attack surface for the agent.
Ingestion points: User messages trigger mandatory skill lookup via the Skill tool.
Boundary markers: None identified in the instructions; the agent is told to "follow it directly."
Capability inventory: The agent is instructed to use the Skill tool, which reads and executes instructions from external files.
Sanitization: There are no instructions for sanitizing or validating the content of the skills being loaded.

using-superpowers