The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local Python scripts to perform specialized searches and integration tests.
Evidence: scripts/omnisearch.py and scripts/test_omnisearch.py utilize subprocess.run to call sibling scripts such as xpoz_search.py and external skill scripts like exa_search.py.
Analysis: Commands are constructed as lists and executed without the shell (shell=False), which effectively prevents shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill aggregates and processes untrusted data from multiple external web and social media sources, presenting a potential surface for indirect injection.
Ingestion points: Data from Brave, Tavily, Xpoz, Exa, and Firecrawl enter the agent context via scripts/omnisearch.py, scripts/brave_search.py, scripts/tavily_search.py, and scripts/xpoz_search.py.
Boundary markers: The skill does not use explicit delimiters or specific instructions to the LLM to ignore potentially malicious instructions embedded in the search results.
Capability inventory: The skill can perform network operations via requests and execute local scripts via subprocess.run.
Sanitization: Content is truncated and normalized for display, but no escaping or filtering for prompt injection patterns is implemented.
[EXTERNAL_DOWNLOADS]: The skill communicates with several external search providers' APIs to retrieve data.
Evidence: Scripts use the requests library to connect to api.search.brave.com, api.tavily.com, and mcp.xpoz.ai.
Analysis: These connections are directed to the official endpoints of the search providers described in the skill's documentation.

omnisearch