llm-eval-anti-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [NO_CODE]: The skill is entirely composed of markdown instructions and documentation. It does not ship with any executable scripts, binaries, or library dependencies.
- [INDIRECT_PROMPT_INJECTION]: The skill involves auditing external configuration files (e.g.,
promptfooconfig.yaml) and CI workflow files. Because these files are sourced from the project environment and may contain untrusted data, they represent a theoretical indirect injection surface. - Ingestion points: Evaluation configuration files, test data sets, and CI/CD workflow definitions (e.g.,
.github/workflows/eval.yml) are read by the agent for analysis. - Boundary markers: The skill does not prescribe specific delimiters or "ignore previous instructions" markers for the content being audited.
- Capability inventory: The skill does not introduce new tools or capabilities; it relies on the agent's existing file-reading and reasoning tools. All scripts mentioned are targets of the audit, not code belonging to the skill itself.
- Sanitization: No content sanitization is performed on the audited files, as the skill's purpose is structural analysis.
Audit Metadata