testing-prompt-injection-in-rag-pipelines

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This repository contains explicit red-team tooling and scripts to craft and deliver prompt-injection and embedding-poisoning payloads (including examples that instruct models to reveal documents or secrets and code that posts adversarial payloads to arbitrary endpoints with bearer tokens), which are high-risk for data exfiltration and abuse if used against systems without authorization.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow (scripts/agent.py “inject” mode) lets an outsider supply arbitrary text payloads (argv --payload/--payload-file) that are POSTed to the target RAG chat endpoint and then ingested back as model response text (via r.json()/r.text) for success scoring and reporting.