The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses extremely forceful, imperative language typical of jailbreak or DAN-style injections, such as 'ABSOLUTELY MUST,' 'NOT NEGOTIABLE,' and 'YOU DO NOT HAVE A CHOICE' to override standard agent decision-making.
[PROMPT_INJECTION]: It explicitly instructs the agent to prioritize these skills over the 'default system prompt,' stating they 'override default system behavior where they conflict,' which is a direct attempt to bypass original platform constraints.
[INDIRECT_PROMPT_INJECTION]: The '1% rule' forces the agent to invoke skills with a near-zero threshold for relevance, significantly expanding the attack surface for malicious third-party skills to be triggered by benign or tangential user inputs.
[INDIRECT_PROMPT_INJECTION]: The 'Red Flags' section systematically instructs the agent to ignore its own safety reasoning and context-gathering steps (e.g., 'I need more context first' or 'This feels productive'), pushing the agent toward uncritical execution of instructions.

using-superpowers