Security audit

Play Dumb

Security checks across malware telemetry and agentic risk

Overview

This is a prompt-only playful mode that intentionally makes the agent act less capable, with no evidence of hidden access, execution, persistence, or data collection.

Install this only if you intentionally want a playful mode that may ask obvious questions, delay answers, or make deliberate mistakes. Avoid enabling it for accuracy-sensitive work, and be aware that broad phrases like “play dumb,” “正常模式,” or “认真点” may change the mode during normal conversation.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The trigger phrases are broad and include common conversational language such as '别那么聪明' and 'play dumb', which can cause the skill to activate unintentionally during normal user interaction. Because this skill intentionally degrades response quality, accidental activation can reduce reliability, cause misunderstanding, and interfere with correct handling of non-exempt tasks even if explicit safety carve-outs exist.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The activation section enumerates multiple ambiguous phrases, including stylistic/persona requests and slang terms, without defining clear boundaries for when the mode starts, how long it persists, or whether it applies only to the next response. In a skill designed to intentionally act less capable, ambiguous invocation materially increases the chance of unwanted persistence and degraded assistance across subsequent turns.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The trigger '检测到 frustration 情绪' is underspecified and relies on subjective emotion detection, which can cause the skill to change behavior unpredictably. In a safety-boundary configuration, ambiguous state transitions are risky because they may fail to deactivate the 'play dumb' mode when the user is distressed, or may toggle behavior inconsistently across languages and contexts.

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The deactivation phrases include common conversational language like '正常模式' and '认真点', which may appear in ordinary dialogue without an intent to control the skill. This can cause unintended mode changes, making the agent's behavior inconsistent and easier to manipulate through prompt wording rather than explicit user consent.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The exit triggers use broad natural-language phrases such as '认真点' and '别装了' that can plausibly appear in ordinary conversation, causing the mode to switch unexpectedly. In a skill that intentionally degrades answer quality, accidental activation or deactivation can undermine reliability and may bypass the intended safety posture at the wrong time.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.