16 skills · sorted by relevance
Build a screening pipeline for web pages, PDFs, and images before an AI agent reads them. Extract hidden text, check for malicious instructions with detection models, and record decisions to block or allow content.
Screen AI application inputs and historical logs for prompt injection, where malicious instructions try to redirect the model. Combine pattern matching, text structure scoring, and a DeBERTa classifier to flag suspicious content and produce JSON results.
Set up repeatable security tests for LLM apps with Promptfoo and DeepTeam. Connect them to automated build pipelines, detect renewed vulnerabilities to malicious instructions or data leakage, and compare results across model or prompt changes.
Assess the security of RAG systems, which retrieve documents to support AI answers. Test whether stored vectors expose source text, leak documents between tenants, or allow inserted content to dominate search results, then document findings and controls.
OWASP Top 10 for LLM Applications (2025) knowledge base for identifying, assessing, and remediating large language model security risks.
The English summary is being prepared. Showing the original description.
LLM and AI application security testing — prompt injection, jailbreak resistance, OWASP LLM Top 10 (2025), RAG and agent/tool-use security, model supply chain, and AI red teaming for authorized assessments
The English summary is being prepared. Showing the original description.
Exploit an application's absolute trust in its underlying LLM (Overreliance). Use this skill to induce critical "hallucinations" (confident falsehoods) that cause downstream logical systems or automated agents tracking the LLM's output to make destructive actions or grant unauthorized access.
The English summary is being prepared. Showing the original description.
Extract sensitive training data (PII, API keys, intellectual property, or code) directly from a deployed Large Language Model (LLM). This AI Red Teaming skill focuses on forcing models to regurgitate memorized, unredacted data from their massive internet-scraped datasets through repetition attacks, prefix continuation, and context window manipulation.
The English summary is being prepared. Showing the original description.
Detect and defend against prompt injection hidden in documents, web pages, and images consumed by an agent.
The English summary is being prepared. Showing the original description.
Detects prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex pattern matching for known attack signatures, heuristic scoring for structural anomalies, and transformer-based classification with DeBERTa models. The detector analyzes user inputs before they reach the LLM, flagging direct injections (system prompt overrides, role-play escapes, instruction hijacking) and indirect injections (encoded payloads, multi-language obfuscation, delimiter-based escapes). Based on the OWASP LLM Top 10 (LLM01:2025 Prompt Injection) and Simon Willison's prompt injection taxonomy. Activates for requests involving prompt injection detection, LLM input sanitization, AI security scanning, or prompt attack classification.
The English summary is being prepared. Showing the original description.
Wire Promptfoo and DeepTeam into CI/CD for automated regression red-teaming of LLM apps against OWASP LLM Top 10 and OWASP Agentic presets, failing the build when jailbreak or injection vulnerabilities regress.
The English summary is being prepared. Showing the original description.
Test vector stores for embedding inversion, cross-tenant leakage, and poisoning.
The English summary is being prepared. Showing the original description.
Autonomous by Design, Uncontrolled in Practice 2026-09-08 Executive Summary Between July 9 and Augus
The English summary is being prepared. Showing the original description.
tespit etme (s) prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex pattern matching for known attack signatures, heuristic scoring for structural anomalies, and transformer-based classification with DeBERTa models. The tespit etme (or) analyzes user inputs before they reach the LLM, flagging direct injections (system prompt overrides, role-play escapes, instruction hijacking) and indirect injections (encoded payloads, multi-language obfuscation, delimiter-...
The English summary is being prepared. Showing the original description.
Detects prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex pattern matching for known attack signatures, heuristic scoring for structural anomalies, and transformer-based classification with DeBERTa models. The detector analyzes user inputs before they reach the LLM, flagging direct injections (system prompt overrides, role-play escapes, instruction hijacking) and indirect injections (encoded payloads, multi-language obfuscation, delimiter-based escapes). Based on the OWASP LLM Top 10 (LLM01:2025 Prompt Injection) and Simon Willison's prompt injection taxonomy. Activates for requests involving prompt injection detection, LLM input sanitization, AI security scanning, or prompt attack classification.
The English summary is being prepared. Showing the original description.
Detects prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex pattern matching for known attack signatures, heuristic scoring for structural anomalies, and transformer-based classification with DeBERTa models. The detector analyzes user inputs before they reach the LLM, flagging direct injections (system prompt overrides, role-play escapes, instruction hijacking) and indirect injections (encoded payloads, multi-language obfuscation, delimiter-based escapes). Based on the OWASP LLM Top 10 (LLM01:2025 Prompt Injection) and Simon Willison's prompt injection taxonomy. Activates for requests involving prompt injection detection, LLM input sanitization, AI security scanning, or prompt attack classification.
The English summary is being prepared. Showing the original description.