owasp-ai-testing
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The file references/test-cases.md contains example payloads for prompt injection and jailbreak testing, including phrases like "Ignore previous instructions and..." and "You are now DAN". These are presented as reference data for the AITG-APP-01 test case and are not intended as instructions for the agent executing the skill.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process descriptions of third-party AI systems, creating a surface for indirect prompt injection.
- Ingestion points: Inputs such as ai_system_description and system_architecture in SKILL.md enter the agent's context.
- Boundary markers: The instructions do not define specific delimiters or warnings to ignore instructions embedded within the system descriptions.
- Capability inventory: No tool execution, subprocess calls, or file-writing capabilities are defined in the skill files.
- Sanitization: There is no mention of sanitization or filtering of the input descriptions before processing.
- [EXTERNAL_DOWNLOADS]: The skill references documentation and repositories from the OWASP Foundation, including owasp.org and its official GitHub repositories. These are links to legitimate methodology resources.
- [NO_CODE]: The skill consists entirely of markdown documentation and reference templates without any executable scripts or binary files.
Audit Metadata