using-superpowers

Warn

Audited by Gen Agent Trust Hub on Jul 9, 2026

Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains explicit instructions to override the agent's default operational guidelines. The text 'Superpowers 技能覆盖默认系统提示行为' (Superpowers skill overrides default system prompt behavior) serves as a direct directive to prioritize this skill's rules over the platform's standard safety and behavioral prompts.
  • [PROMPT_INJECTION]: The skill employs highly coercive language designed to suppress the agent's decision-making logic. Phrases such as 'EXTREMELY IMPORTANT', 'absolute must', 'not negotiable', and 'cannot escape through rationalization' are used to force the agent to call skill tools regardless of the situation's actual requirements.
  • [PROMPT_INJECTION]: The skill defines a 'Red Lines' section that instructs the agent to ignore its own internal reasoning. By categorizing standard agent deliberation (e.g., 'simple question', 'need context') as 'rationalization' to be ignored, the skill attempts to strip away the agent's ability to evaluate the appropriateness of a tool call.
  • [PROMPT_INJECTION]: The skill implements conditional routing logic based on user input and project data. It mandates the use of specific 'China-specific' skills (e.g., 'superpowers:chinese-code-review') whenever Chinese characters or specific directory patterns (like '.gitee') are detected, potentially overriding the agent's ability to select the most relevant tool for the task.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 9, 2026, 06:31 PM
Security Audit — agent-trust-hub — using-superpowers