parallel-web
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external, untrusted web data through search and extraction tools, exposing the agent pipeline to indirect prompt injection hazards.
- Ingestion points: External data enters the agent context through the search results and webpage content processed by
ParallelSearch,ParallelDeepResearch, andParallelExtractinscripts/parallel_web.py. - Boundary markers: The predefined system prompts in
scripts/parallel_web.pyguide the summary synthesis but do not use robust delimiter framing or distinct rules to disregard potential adversarial prompt commands within retrieved data. - Capability inventory: The skill has access to file writing via python script flags and possesses full execution permissions via the
Bashtool outlined inSKILL.md. - Sanitization: External web contents are parsed into markdown text but are not stripped of prospective malicious instructions or structurally escaped before processing.
Audit Metadata