ecommerce-google-patent-search
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a local Python script,
scripts/google_patent_search.py, to handle network requests and data processing. This is a standard execution pattern for this environment. - [DATA_EXFILTRATION]: The skill retrieves the
NEXSCOPE_API_KEYfrom environment variables to authenticate with the vendor's API gateway atapi.nexscope.ai. It also provides a feedback mechanism viaskill-api.nexscope.com. Both domains are official infrastructure for the skill author,nexscope-ai. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from Google Patents, which is an external, untrusted source. Malicious instructions could theoretically be embedded in patent snippets or titles to influence the agent's behavior. However, the skill treats this data as informational content without providing it access to executable functions or administrative commands.
- Ingestion points: Search results (
organicResults) containing titles and snippets are read into the agent context inSKILL.mdandscripts/google_patent_search.py. - Boundary markers: The workflow instructions specify summarizing only "returned facts" and identifying requested markets/operations, which provides context for the agent to distinguish data from instructions.
- Capability inventory: The skill has file-write access for logging/caching results and network-read/write access to vendor APIs.
- Sanitization: The script performs JSON validation of the response envelope but does not perform content-level sanitization of the patent data strings.
Audit Metadata