react-flow-architect
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines logic for processing and rendering graph data which may originate from untrusted sources.
- Ingestion points: The
buildVisibleGraphandsearchNodesfunctions inreferences/detailed-guide.mdprocess node arrays and search query strings. - Boundary markers: The guide includes a mandatory instruction to "render labels as text and never trust imported HTML/URLs," providing a boundary against cross-site scripting (XSS) or prompt injection via rendered data.
- Capability inventory: The skill is limited to React component logic and standard graph layout algorithms; it does not request file system access, network capabilities, or shell execution.
- Sanitization: The documentation provides manual guidance on treating data as plain text to prevent exploitation of the UI rendering layer.
- [METADATA_POISONING]: The skill frontmatter includes a
risk: criticaltag which is not supported by the actual functional content of the skill. - Evidence:
risk: criticalinSKILL.mdfrontmatter. - Analysis: While the metadata claim is inaccurate relative to the benign code provided, it does not introduce a security vulnerability into the agent environment.
Audit Metadata