internet-lookup-verifier

The provided files, SKILL.md and _meta.json, are purely descriptive and contain no executable code, scripts, or direct commands for the AI agent to execute.

• SKILL.md: This Markdown file describes the purpose, functionality, usage, and limitations of the 'Internet Lookup Verifier' skill. It mentions the use of web_search and web_fetch tools, which are standard capabilities for an AI agent to interact with the internet. However, this file does not contain the implementation of these tools or any direct instructions that could lead to prompt injection, data exfiltration, privilege escalation, or other malicious activities. The front matter and content are benign.
• _meta.json: This JSON file contains metadata about the skill, such as owner, display name, and a commit reference. All fields are standard and do not contain any malicious patterns or obfuscated content. The GitHub commit URL is a reference and does not trigger any download or execution.

Threat Category Analysis:

• Prompt Injection: No patterns detected. The skill's description of how it triggers based on keywords is not a prompt injection attempt itself.
• Data Exfiltration: No sensitive file paths or explicit network operations for exfiltration are present.
• Obfuscation: No obfuscation techniques (Base64, zero-width characters, homoglyphs, URL/hex/HTML encoding) were found.
• Unverifiable Dependencies: No external dependencies are declared or installed by these files.
• Privilege Escalation: No commands indicating privilege escalation were found.
• Persistence Mechanisms: No patterns for establishing persistence were found.
• Metadata Poisoning: No malicious instructions were found in the metadata fields of either file.
• Indirect Prompt Injection: While the nature of a skill that performs web lookups (using web_search and web_fetch) makes it inherently susceptible to indirect prompt injection from the content it retrieves, the provided files themselves do not contain such an injection. This is a general risk for such skills, not a specific finding in these files.
• Time-Delayed / Conditional Attacks: No conditional logic for delayed attacks was found.

Given that these files are purely descriptive and do not contain any executable code or direct instructions, they are considered safe. The analysis is based solely on the provided content.