Security audit
Security checks for vulnerabilities and agentic risk
This is a simple help-card skill that displays caveman-mode reference text and does not install code, persist changes, or access data.
Before installing, note that ordinary phrases like "caveman help" or "normal mode" may interact with caveman-mode behavior. This is a usability concern, not evidence of unsafe access or hidden behavior.
The skill advertises broad natural-language triggers such as "caveman help" and "what caveman commands," which can be matched during ordinary conversation rather than explicit invocation. That increases the chance of unintended activation, causing the agent to switch behavior and emit skill content when the user did not actually intend to invoke the reference card.
The deactivation phrases "stop caveman" and especially "normal mode" are generic enough to appear in unrelated dialogue, quoted text, or instructions about other systems. This can unintentionally disable the mode, creating reliability and control issues and making the skill susceptible to prompt-injection-style interference through casual language.
The instruction "Output in caveman style" imposes a specific language/style transformation by default rather than explicitly offering it as an opt-in at response time. This can conflict with language or locale policy expectations when a user requests help content but does not affirmatively choose that style in the current interaction.
No suspicious patterns detected.
