compliance-audit
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill represents a vulnerability surface for indirect prompt injection because it is designed to ingest and analyze untrusted external data.
- Ingestion points: The skill uses the
WebFetchtool to retrieve HTML source code, meta tags, and headers from arbitrary URLs provided by the user (SKILL.md, Phase 1: Detection Scan). - Boundary markers: There are no explicit instructions or delimiters mentioned to tell the agent to ignore natural language instructions found within the scraped HTML or meta tags.
- Capability inventory: The skill possesses significant capabilities, including the ability to read local files (
Read,Glob,Grep) and perform network operations (WebFetch). - Sanitization: The instructions do not specify any sanitization or filtering of the content retrieved from external websites before processing.
- [COMMAND_EXECUTION]: The skill uses
GlobandGrepto perform automated searches across a local codebase for patterns related to compliance (e.g., searching for authentication patterns, encryption, or payment processing). While these are standard tools for analysis, they involve automated file system interaction based on high-level instructions.
Audit Metadata