The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the tooluniverse Python package from a public registry as documented in SKILL.md and references/installation.md (uv pip install tooluniverse).
[COMMAND_EXECUTION]: The skill utilizes a central run() method to execute a large catalog of 600+ scientific tools, including machine learning models and APIs. This provides a broad functional surface for the agent to execute computational research tasks as shown in references/tool-execution.md.
[PROMPT_INJECTION]: The skill is designed to ingest and process data from external scientific repositories such as PubMed, UniProt, and GEO. As demonstrated in scripts/example_workflow.py and references/tool-composition.md, this untrusted data (e.g., sequences, abstracts) is passed to other tools or LLMs for summarization, creating a surface for indirect prompt injection.
Ingestion points: UniProt_get_sequence, PubMed_search, and GEO_download_dataset functions in scripts/example_workflow.py.
Boundary markers: The provided examples do not include explicit delimiters or instructions to ignore embedded commands in the scientific data.
Capability inventory: The skill has access to 600+ tools via tu.run() and can summarize content using LLM_summarize.
Sanitization: There is no evidence of specific sanitization or filtering for the external content before it is processed by subsequent tools.

tooluniverse