Security audit
Security checks for vulnerabilities and agentic risk
This skill does not show malware or data theft, but it broadly and silently changes agent reasoning behavior and output formatting with unclear thresholds.
Review before installing. This skill is not destructive and does not request data access, but it can silently change how the agent reasons and appends markers to responses across unrelated messages. Install only if you want that global behavior and are comfortable with the unclear activation thresholds.
Triggering on every user message is overly broad for a skill that can alter reasoning mode and output format. A global pre-processing hook increases the blast radius of any bad instruction in the skill and can cause persistent, unintended behavior across unrelated conversations.
Although the rubric provides some scoring guidance, the actual activation thresholds are partially garbled and remain ambiguous, leaving room for inconsistent or excessive activation. Ambiguous autonomous behavior is risky because implementations may default to more invasive reasoning-state changes or visible markers without clear justification.
Mandating visual markers in all qualifying responses changes user-facing output without consent and leaks internal processing state. This can confuse users, interfere with downstream integrations that expect clean text, and create a covert signaling channel controlled by the skill.
The skill claims to be only mental preprocessing, but it also instructs the agent to issue internal commands or tool calls that alter session reasoning state and to append visible markers to outputs. That mismatch can cause hidden state changes and user-visible behavior outside the declared scope of the skill, reducing transparency and increasing the chance of policy or UX violations.
The file states 'No external tools needed' while explicitly directing use of a tool or command to activate reasoning. This contradiction undermines operator trust and can bypass safeguards that rely on accurate declarations of tool usage and side effects.
No suspicious patterns detected.
