NYC

physics-rendering-expert

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [Indirect Prompt Injection] (LOW): The skill manifest grants the agent high-privilege tools (Bash, Write, Edit) alongside web-search and fetching capabilities (Firecrawl, WebFetch). This configuration establishes an indirect prompt injection surface where instructions embedded in untrusted web content could potentially manipulate the agent into executing malicious commands.
  • Ingestion points: mcp__firecrawl__firecrawl_search, WebFetch tools used for data retrieval.
  • Boundary markers: Not explicitly defined in the skill instructions to separate untrusted data from the reasoning context.
  • Capability inventory: Bash (shell access), Write (filesystem access), Edit (filesystem modification).
  • Sanitization: No explicit sanitization or validation protocols are mentioned for content retrieved via WebFetch.
  • [Remote Code Execution] (SAFE): No remote scripts or unverified dependencies are downloaded or executed. The code snippets provided are illustrative C++ implementations for physics logic.
  • [Data Exposure & Exfiltration] (SAFE): No sensitive file paths, hardcoded credentials, or suspicious exfiltration endpoints were identified.
  • [Obfuscation] (SAFE): No attempts to hide code using Base64, zero-width characters, or homoglyphs were detected. The use of HTML entities in the changelog is standard for MDX compatibility.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 05:55 PM