The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize information from a wide variety of untrusted external web sources, including search results, social media feeds, and academic repositories. This creates a significant attack surface for indirect prompt injection, where malicious instructions embedded in search results or web pages could influence the agent's behavior during report generation or task execution.
[DYNAMIC_EXECUTION]: The skill employs dynamic loading of external modules and path manipulation. Specifically, multi_model_intelligence.py attempts to dynamically import the gemini_cli_skill module at runtime. Furthermore, enhanced_research_agent.py and scripts/run_deep_research.py modify the system path to import modules from distant parent directories. These techniques can be used to load unauthorized code and evade static analysis.
[REMOTE_CODE_EXECUTION]: The script scripts/run_deep_research.py has a dependency on an unverified external package called codex-rs. The skill attempts to import multiple components from this package to perform its core search and research pipelines, posing a supply chain risk as the source and integrity of this package are not verified within the skill's own context.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of several well-known Python packages for web scraping and data processing, including requests, beautifulsoup4, aiohttp, lxml, fuzzywuzzy, and python-levenshtein. While these are established libraries, the total dependency footprint, including unverified local modules, increases the overall risk profile.

web-search-deepresearch