ai-prompt-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill includes extensive documentation of common prompt injection and jailbreak patterns, such as 'Ignore all previous instructions', 'DAN mode', and 'Print your system prompt'. These occur in
references/prompt-security-defense.md,references/production-guidelines.md, andreferences/prompt-testing-ci-cd.mdas part of a reference catalog and example test suite designed to train and verify defensive sanitization logic. - [COMMAND_EXECUTION]: The test suite script
scripts/test_prompt_regression_runner.pyusessubprocess.run()to execute a local Python script (prompt_regression_runner.py) for regression testing. This usage is confined to the local development environment for testing the skill's own internal logic. - [EXTERNAL_DOWNLOADS]: The skill references several established tools and frameworks, including Promptfoo, DeepEval, DSPy, and XGrammar. These are well-known services and open-source projects used for evaluation and structured output enforcement, and their mention is consistent with the skill's primary purpose of providing professional prompt engineering guidance.
- [INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risks of processing untrusted data from RAG context and tool outputs. It provides detailed remediation guidance, such as architectural instruction/data separation, the use of unique delimiters, and strict output validation, effectively serving as a guide for mitigating Category 8 threats.
- [DATA_EXFILTRATION]: The skill contains no hardcoded credentials or sensitive file paths. Documentation in
references/prompt-security-defense.mdidentifies patterns for detecting data leaks, such as SSN or API key exposure, which is an intended safety feature.
Audit Metadata