a11y-violation-gate
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from accessibility scanners (e.g., axe-core, pa11y, Lighthouse) that analyze external web content. This data, which includes URLs and element selectors, is interpolated into the agent's context through console output. An attacker-controlled web page could theoretically embed malicious instructions in these fields to influence the agent's subsequent logic during report generation or baseline evaluation.
- Ingestion points: Scanner results from files such as
axe-results.jsonand various reports within thereports/directory are parsed by the gate scripts inSKILL.mdand the evaluation fixtures. - Boundary markers: The skill does not implement boundary markers or instructions to isolate scanner-derived content, allowing the agent to potentially interpret it as commands.
- Capability inventory: The provided scripts are limited to local file system read operations and standard output; they do not perform network requests, write files, or execute shell commands directly.
- Sanitization: Findings such as element selectors, rule IDs, and page URLs are printed to the console as raw strings without validation or escaping.
Audit Metadata