web-perimeter-owasp-2025-audit
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill mentions an optional audit tool ('@stealthmark/cli') which can be executed via npx for public domain verification. This involves downloading and running code from a remote registry.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from HTTP response headers and bodies during audits. It implements boundary markers by instructing the agent to use specific regex signatures for detection and to report only boolean signals rather than raw content. The ingestion points are curl outputs in SKILL.md. Capability inventory includes network requests via curl, shell command execution, and local file storage. Sanitization is addressed by limiting findings to defined patterns.
- [DYNAMIC_EXECUTION]: Provides a local Python script fixture in SKILL.md to simulate an edge environment for testing detection logic. This script is a static template used for offline validation.
- [COMMAND_EXECUTION]: Employs curl for remote probing and standard shell utilities for managing temporary directories and generating cache-buster strings.
Audit Metadata