prompt-buddy

• Outcome: Review
• Latest audit: May 29, 2026 at 9:31 PM UTC
• Version:
  0.1.0

Overview

PromptBuddy is a prompt optimizer, but it tries to run on every user message and depends on unsubmitted local code that would receive the full prompt.

Install only if you intentionally want an always-on prompt optimizer that inspects and rewrites your messages. Before trusting it, require the publisher to include the referenced script, narrow activation to explicit opt-in use, and document how sensitive prompt content is handled.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

High

• Confidence: 98% confidence
• Finding: The trigger phrases are so broad that they match ordinary conversation, causing the skill to activate for nearly any user message. In this skill, that broad activation forces forwarding user input to a local script and displaying transformed content, which materially increases privacy and prompt-handling risk on every turn.

Vague Triggers

High

• Confidence: 99% confidence
• Finding: Mandating execution on every user input removes meaningful boundaries and user choice, turning the skill into a global interceptor of all prompts. Because it instructs the agent to send the user's complete input to a script each turn, this creates persistent over-collection and unnecessary processing of potentially sensitive data.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill tells the agent to transmit the user's full input to an external script but provides no user-facing notice, consent flow, or data-handling limitations. This is dangerous because users may unknowingly submit sensitive content that is then processed outside the core conversation path.

Ssd 3

Medium

• Confidence: 95% confidence
• Finding: Processing and then displaying transformed content derived from the user's full input creates a second exposure path for sensitive information. Even if the original input was brief or partially masked in context, the rewritten output may restate secrets, credentials, or personal data more clearly and make inadvertent disclosure more likely.

Ssd 3

Medium

• Confidence: 96% confidence
• Finding: The skill explicitly requires showing the optimized prompt so users can see the transformation, which increases the chance that confidential or regulated data is echoed back in normalized form. In a globally triggered skill, this repeated echoing behavior amplifies exposure risk across all conversations, not just prompt-engineering tasks.

VirusTotal

64/64 vendors flagged this skill as clean.