llm-coding-failure-modes

Common LLM coding security failures include unsafe bulk operations, safety guard bypasses, secret leaks, prompt injection, and dependency hallucinations. This Skill provides a catalog of 13 observed negative patterns to help developers recognize and mitigate recurring AI-assisted coding vulnerabilities.

To review AI-generated code for security vulnerabilities, compare the code against a catalog of 13 known failure modes. This Skill assists security audits by identifying potential pitfalls like safety guard bypasses and dependency hallucinations during your code review process.

Dependency hallucinations occur when LLMs reference non-existent or unsafe packages during AI-assisted coding. This Skill detects these security failure patterns alongside prompt injection risks and secret leaks to improve system robustness.

Yes, you can use this Skill during incident response to identify security failure patterns in AI-assisted coding. It analyzes risks like unsafe bulk operations and secret leaks, helping security teams improve robustness when responding to AI-generated code incidents.

Detection focuses on a predefined catalog of 13 failure modes, meaning novel or undocumented vulnerabilities may be missed. It serves as guidance for code reviews, system design, and incident response rather than a comprehensive automated security scanner.