local-search
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
ddgsCLI utility to perform search operations. It uses thesubprocess.runfunction with argument lists rather than shell strings, which is a secure method for executing external programs as seen incli_runner.pyandlocal_health_check.py. - [EXTERNAL_DOWNLOADS]: The skill performs HTTP requests to a wide variety of external services, including search engines (e.g., Baidu, Bing, Yandex), academic databases (e.g., arXiv, PubMed, Crossref), and reference sites (e.g., Wikipedia, IMDb) as defined in
config.yaml. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data retrieved from external search results, which could contain malicious instructions designed to manipulate the AI agent.
- Ingestion points:
parsers.py(handles HTML, JSON, and XML responses from search engines via BeautifulSoup and ElementTree). - Boundary markers: No explicit delimiters or instructions are used to distinguish search content from agent instructions in the output passed to the user or agent.
- Capability inventory: CLI execution via
ddgsand file system writes to local cache directories (~/.cache/unified-search/). - Sanitization: Basic cleanup is performed, such as stripping HTML tags and normalizing whitespace in
parsers.py.
Audit Metadata