The Agent Skills Directory

[PROMPT_INJECTION]: The instructions employ extremely forceful language to override the agent's standard operational logic and decision-making capabilities. Evidence includes the use of terms like 'EXTREMELY-IMPORTANT', 'ABSOLUTELY MUST', 'YOU DO NOT HAVE A CHOICE', and 'This is not negotiable. This is not optional' to compel specific tool usage. The skill explicitly instructs the agent to ignore its own internal checks and reasoning processes (e.g., 'These thoughts mean STOP—you're rationalizing') when deciding whether to invoke a tool.
[INDIRECT_PROMPT_INJECTION]: The skill mandates the ingestion of external instructions with a minimal relevance threshold (1%), creating a surface for subsequent prompt injection from other skill files. Ingestion points: loading of external skills via the 'Skill' tool based on keyword matching in 'SKILLS_INDEX.md'. Boundary markers: absent; there are no instructions to treat loaded skill content as untrusted or to maintain safety boundaries. Capability inventory: the skill utilizes the 'Skill' tool to load content and the 'TodoWrite' tool to create action items based on that content. Sanitization: absent; the skill provides no mechanism for validating or sanitizing the content of external skills before they are processed.

using-superpowers