Security audit

president.skill — Think and Act Like an American President

Security checks for vulnerabilities and agentic risk

Overview

This is an instruction-only presidential roleplay skill with no code or data access, but its broad triggers may unexpectedly switch conversations into a Trump-style persona.

Install this only if you want a stylized presidential, currently Trump-centered, decision and roleplay perspective. Use explicit prompts when you want that mode, and say "exit," "stop," or "return to normal" if it activates unexpectedly. For legal, financial, medical, technical, or politically sensitive decisions, ask for neutral analysis instead of relying on the persona framing.

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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 (11)

Vague Triggers

High

Confidence: 97% confidence
Finding: The trigger list is extremely broad and includes common names, slogans, generic questions, and conversational slang that can appear in ordinary user input without an explicit request to activate the skill. This creates prompt/skill injection-by-accident risk, where the agent may silently enter a politically loaded persona and change its behavior, tone, or recommendations in contexts unrelated to the user's actual intent.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The default activation rule ('what would a president do' => Trump by default) is ambiguous because it maps a generic request for a presidential perspective to a specific persona without clear user consent. In practice, this can cause unintended role activation, biased outputs, and inconsistent behavior, especially in sensitive decision-support contexts where persona choice materially affects advice.

Vague Triggers

High

Confidence: 96% confidence
Finding: The trigger list is extremely broad, including generic terms like "president," any president's name, nickname, or famous slogan. This makes accidental activation likely when a user is merely discussing history, politics, or quoting public figures, causing unintended persona switching and instruction injection into unrelated conversations. In this context, the roleplay nature of the skill increases risk because broad matching can silently override normal assistant behavior.

Vague Triggers

High

Confidence: 94% confidence
Finding: The activation logic says that if input "contains trigger word," the system should identify and activate a president, but it does not distinguish mention from request. A user could reference Trump, Lincoln, or a slogan in analysis, criticism, or quoting, and the skill would still activate, creating prompt-routing ambiguity and unintended behavior changes. Because this is framed as automatic activation, the risk is higher than a mere documentation issue.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: Defaulting generic requests like "what would a president think" to a specific political figure creates an underspecified and potentially manipulative trigger path. It can steer users into an unintended persona without informed choice, especially when multiple perspectives are supposedly available. The risk is somewhat lower than the broad trigger issue but still material because it hardcodes a controversial default into ambiguous requests.

Natural-Language Policy Violations

Medium

Confidence: 91% confidence
Finding: The skill instructs the assistant to respond in first person and imitate a specific public figure's tone and catchphrases without requiring explicit user opt-in at activation time. This can mislead users about authorship, intensify accidental activation harm, and make it harder to maintain clear boundaries between normal assistant behavior and persona mode. In combination with the broad triggers elsewhere in the file, this becomes more dangerous because users may be placed into immersive roleplay unexpectedly.

Vague Triggers

High

Confidence: 95% confidence
Finding: The trigger conditions are broad enough to activate on ordinary negotiation or complaint language, which can cause unintended persona switching without clear user consent. In this skill, that matters more because activation leads directly into a strong political roleplay persona with first-person identity emulation, increasing the chance of misleading, manipulative, or off-policy responses in unrelated contexts.

Natural-Language Policy Violations

Medium

Confidence: 98% confidence
Finding: The skill explicitly instructs the agent to speak as Donald J. Trump in first person and to stay in character, while giving only a one-time disclaimer. That creates a meaningful impersonation risk: users may miss or forget the disclaimer, and the instruction not to break character makes it harder for the system to maintain clear separation between simulation and real identity.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The trigger words include bare numeric activators like "45" and "47", which are highly ambiguous and likely to appear in ordinary conversation, dates, rankings, addresses, or counts. In an auto-activation system, this can cause accidental invocation of the Trump persona, leading to unintended style/persona switching and possible policy or behavioral drift in unrelated contexts.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The phrase "Trump would" is overly broad and can naturally occur in many comparative, historical, political, or hypothetical discussions without intending to invoke a skill. Because this skill is a high-priority presidential persona with strong behavioral framing, accidental activation could redirect responses into a partisan or persona-driven mode unexpectedly.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The phrase "if Trump were here" is a common conversational construction that may appear in benign discussion, debate, or writing prompts. Its breadth increases the chance of accidental skill activation, and in this skill's context that means unexpectedly shifting into a named political persona, which can affect tone, judgment framing, and output safety in unrelated tasks.

Static analysis

No suspicious patterns detected.