The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing several third-party research libraries and cloning repositories from GitHub, including gpt-researcher, deepsearcher, local-deep-research, DeepResearch (Alibaba NLP), and khoj. These references point to widely recognized open-source projects in the AI research ecosystem.
[COMMAND_EXECUTION]: The guides contain standard shell commands for environment setup and tool invocation, including pip install, git clone, docker run, and REST API interactions via curl. These are presented as baseline requirements for using the described research tools.
[DYNAMIC_EXECUTION]: The kosmos-scientist-guide/SKILL.md describes a workflow where an AI Scientist agent autonomously implements and executes Python code for scientific experiments, training loops, and data analysis. This is a deliberate functional feature of the autonomous discovery frameworks documented in the skill.
[INDIRECT_PROMPT_INJECTION]: The research agents described (e.g., OpenResearcher, GPT Researcher) are designed to ingest untrusted data from the web and academic papers to generate reports. This creates an attack surface for indirect prompt injection if external documents contain adversarial instructions. Ingestion points: Web search results (Tavily, Serper, Bing), academic databases (OpenAlex, arXiv, PubMed), and local PDF/Markdown document collections. Boundary markers: The high-level usage guides do not explicitly define specific delimiter strategies for isolating ingested content. Capability inventory: These research tools typically have the ability to perform network requests, write files (reports), and in some cases, execute code for analysis. Sanitization: Specific input sanitization logic is not detailed within the provided documentation snippets.

deep-research-skills