qa-browser-automation
Fail
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: HIGHINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML content from external websites via the
scripts/accessibility_auditor.pyscript. The tool's output includes strings extracted from the audited pages, such as link text, alt text, and ARIA labels. This creates a surface for indirect prompt injection, where an attacker-controlled website could embed malicious instructions in these attributes to influence the agent's actions during the audit process. - Ingestion points: External HTML content processed by
scripts/accessibility_auditor.pyfrom local files or stdin. - Boundary markers: The skill's instructions do not include boundary markers or delimiters to warn the agent about potential instructions embedded within the auditing tool's output.
- Capability inventory: The agent has the capability to write files (session data, baselines, reports), perform git commits, and automate browser interactions via Chrome MCP.
- Sanitization: The auditing script extracts raw text from HTML attributes and includes it in its findings without sanitization or escaping to prevent the agent from interpreting the content as instructions.
- [DYNAMIC_EXECUTION]: The script
scripts/test_report_generator.pyuses theimportliblibrary andexec_moduleto dynamically load sibling Python scripts (such asqa_health_scorer.py) at runtime. While the implementation currently targets local files using hardcoded strings, the pattern of dynamic module loading from computed file paths is a high-privilege operation that would represent a significant risk if the module name resolution were ever exposed to external or untrusted input.
Recommendations
- HIGH: Downloads and executes remote code from: https://example.com - DO NOT USE without thorough review
Audit Metadata