Scan untrusted text for 16 categories of prompt injection attacks.
npx skills add https://github.com/dexhunter/Logi-Lobsterism --skill input-guard-dexhunter
Please help me install this Agent Skill. Skill: input-guard Source: https://github.com/dexhunter/Logi-Lobsterism/tree/main/skills/input-guard Command: npx skills add https://github.com/dexhunter/Logi-Lobsterism --skill input-guard-dexhunter
This Skill acts as a crucial security layer, scanning untrusted external text to detect and block prompt injection attacks before they can compromise an AI agent.
Use the input-guard skill to scan the following text for prompt injection threats.
High-intent search queries and answers about installing and using this skill.
Scanning external text for prompt injection attacks detects malicious instructions before agent processing. Input Guard identifies 16 threat categories, including instruction override and data exfiltration, to block harmful content from compromising your AI.
Prompt injection is an attack where untrusted external text contains embedded malicious instructions. Input Guard scans for these threats, preventing role manipulation and data exfiltration by blocking malicious content before your AI agent processes it.
Scanning for prompt injection threats supports multiple languages including English, Korean, Japanese, and Chinese. Input Guard detects malicious instructions across these languages to prevent cross-lingual attacks from compromising your AI agent.
Configuring sensitivity levels for threat detection allows you to adjust scanning strictness for API responses. Input Guard provides configurable sensitivity levels and optional LLM-powered analysis to identify prompt injection threats accurately.
Input Guard works with social media posts, web articles, and API responses by scanning the untrusted text for threats. If a prompt injection is detected, the content is blocked and an alert is sent to prevent harm.
Rule-based detection limitations include missing novel or complex prompt injection patterns. Input Guard mitigates this by offering optional LLM-powered analysis alongside its 16-category threat detection to catch sophisticated instruction override attempts.
mhdxbilalApply security framing to untrusted external content for safe analysis.
sandmark78Detect dangerous commands and prompt injections in user-provided text.
phatblatAutomate web application security assessments with reconnaissance, threat modeling, and vulnerability analysis.
gula00Scan AI agent skills and inputs for injection and exfiltration threats.
