Detects and tags planted instructions in untrusted text to prevent AI hijacking.
npx skills add https://github.com/JGalego/Bag-of-Tricks --skill mole-jgalego
Please help me install this Agent Skill. Skill: mole Source: https://github.com/JGalego/Bag-of-Tricks/tree/main/mole Command: npx skills add https://github.com/JGalego/Bag-of-Tricks --skill mole-jgalego
This Skill identifies and tags potential harmful instructions hidden in untrusted text, preventing them from influencing the AI's processing.
Run the 'mole' skill on a text snippet to detect planted instructions.
High-intent search queries and answers about installing and using this skill.
Identify prompt injection by scanning untrusted text for planted instructions such as role spoofing, persona jailbreaks, and instruction overrides to prevent AI hijacking at the input stage.
A persona jailbreak is a planted instruction within untrusted text that attempts to override the AI's assigned role, hijacking behavior by forcing it to adopt an unauthorized persona.
To scan a document for hidden instructions and role spoofing, run an input-side security detection process that tags instruction overrides and prompt-leak attempts before AI processing.
Input-side security detection works for preventing instruction overrides by operating at the text input stage to identify and tag harmful commands before they influence AI processing.
The best way to prevent AI hijacking from prompt-leak attempts is to scan untrusted text inputs to identify and tag planted instructions before the AI processes the content.
Detecting planted instructions via text analysis is limited to identifying known patterns like role spoofing and instruction overrides at the input stage, and may miss novel or obfuscated prompt injection techniques.
thejefflarsonDetect and fix prompt construction patterns that allow user input to influence system prompts.
8DionysusRedact sensitive details from technical material and place sanitized output in a public-safe location.
latestaiagentsValidate and sanitize user inputs to defend against prompt injection attacks.
ricableScan codebases for CVEs, secrets, and AI manipulation threats.
