LLM application security anti-patterns reference loaded by super-review:run

What problem does it solve?

Prevents prompt injection, unsafe tool use, and LLM-driven security regressions by guiding reviewers to spot high-risk anti-patterns in LLM and RAG code before they ship.

Core Features & Use Cases

• Anti-pattern detection for LLM apps: Covers direct/indirect prompt injection, untrusted context handling, and instruction leakage risks.
• Unsafe output-sink tracing: Identifies when model output is routed into eval, shell/SQL execution, redirects, or dangerous HTML rendering without validation.
• Tool-call and agency hardening: Flags missing tool-argument schema validation, excessive agent permissions, unbounded generation cost risks, and unsafe streaming behavior.
• RAG and supply-chain risks: Targets vector-store retrieval poisoning and slopsquatted or hallucinated dependency patterns tied to LLM workflows.

Quick Start

Use this reference when reviewing a pull request that includes LLM SDK calls or RAG/vector-store code to systematically flag the most critical security anti-patterns.