skills.rest
nvtruongops
nvtruongopsCOMMUNITY
Explore all skills by this author →
GitHub

llm-security-research

Guides prompt injection and jailbreak research with curated academic references and citation standards.

1|Updated Sep 1, 2026
Search Tags:#owasp#citations#literature review#guardrail#jailbreak#prompt injection#llm security
One-click install
npx skills add https://github.com/nvtruongops/pi-guard --skill llm-security-research-nvtruongops
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: llm-security-research
Source: https://github.com/nvtruongops/pi-guard/tree/main/.agents/skills/llm-security-research
Command: npx skills add https://github.com/nvtruongops/pi-guard --skill llm-security-research-nvtruongops

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Conducting rigorous academic research on LLM security requires finding post-2022 papers on prompt injection and jailbreaks, classifying attacks against OWASP LLM01 taxonomy, and maintaining consistent citations without paywalled or unsupported claims. ## Core Features & Use Cases - Local References Archive: Provides a catalog of 18 verified papers (DeBERTa-v3, Llama Guard, NeMo Guardrails, GCG attacks) mapped to thesis chapters with BibTeX metadata. - Threat Taxonomy Guidance: Classifies attacks into direct injection, indirect injection, jailbreaks, and obfuscation techniques using a 5D threat analysis framework. - Citation Integrity Workflow: Enforces dual-linking (DOI plus open-access PDF), on-page anchor standards, and a >= 2022 publication rule for all literature. - Use Case: When writing a thesis chapter on jailbreak defenses, check REFERENCES_LOG.md first, reuse archived papers, and verify new sources with the URL validation script. ## Quick Start Ask the assistant to find and cite open-access papers from 2022 or later about indirect prompt injection defenses for the PI-Guard thesis.

Frequently Asked Questions about llm-security-research

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I research prompt injection attacks for an academic paper?▼

Start by checking the local REFERENCES_LOG.md archive of 18 verified papers covering direct injection, indirect injection, and jailbreaks. Classify findings using the OWASP LLM01 taxonomy and only cite works published from 2022 onward.

What are the main types of prompt injection attacks?▼

The taxonomy covers direct prompt injection (system prompt override), indirect prompt injection (payloads in retrieved data), jailbreaks (persona, hypothetical, cognitive overload), and obfuscation techniques like leetspeak, Base64 encoding, and spaced text.

How do I find open-access PDFs for paywalled security papers?▼

Use the verify_resource_url.py script with a DOI to automatically resolve open-access PDF links from OpenAlex or Semantic Scholar. Every citation must include both the official DOI and a free full-text link such as arXiv.

Which guardrail baselines should I compare against in LLM security research?▼

The archived references include Meta Llama Guard, NVIDIA NeMo Guardrails, ProtectAI DeBERTa classifiers, and SmoothLLM randomized smoothing. These serve as SOTA baselines for benchmarking prompt injection detection systems.

Can I cite papers published before 2022 in LLM security research?▼

No, the project enforces a strict rule that all scientific papers must be published from 2022 onward, covering the modern InstructGPT and post-ChatGPT era. One exception exists for foundational works like Saltzer 1975 on protection principles.

Related Skills

Explore More Education & Research Skills→
jpologjpolog

literature-retrieval

Locate, download, and organize open-access academic papers into a literature library.

Community
Advanced
woaijiadanoowoaijiadanoo

thesis-pipeline

Orchestrate LaTeX thesis chapters through latex-guard, academic-polisher, and optional humanizer.

Community
Advanced
hanuminhanumin

research-verify

Validate research citations, DOI links, and formatting consistency in scientific reports.

Community
Advanced
Agents4Academia-AIAgents4Academia-AI
9

zotero

Extract Zotero PDF text, highlights, and ink annotations into structured Markdown files.

Official
Advanced