adversarial-verifier
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill serves a purely defensive and analytical purpose, providing instructional frameworks for identifying bugs and logic flaws. All external references are to well-known academic or industry research papers from Meta AI and Anthropic.
- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by design, as it is intended to process and analyze untrusted third-party artifacts such as code snippets, data sets, and architecture documents.
- Ingestion points: External artifacts are identified and gathered during the 'Activation Protocol' phase described in
SKILL.md. - Boundary markers: The instructions do not explicitly mandate the use of delimiters or protective tagging when interpolating untrusted data into the verification prompts.
- Capability inventory: The skill leverages the host agent's native capabilities to read files and analyze logic, which could potentially be targeted by instructions hidden within the artifacts being audited.
- Sanitization: The protocol does not include specific steps for sanitizing or filtering untrusted content before the agent processes it.
Audit Metadata