The Agent Skills Directory

[SAFE]: The skill implements a robust, multi-phase clinical workflow for variant interpretation, adhering to ACMG/AMP 2015 guidelines. It integrates various professional bioinformatics databases and state-of-the-art predictive models (AlphaFold2, AlphaMissense, SpliceAI, etc.).\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including PubMed abstracts, preprint servers (BioRxiv, MedRxiv), and community-curated databases like ClinVar. While this creates a theoretical surface for indirect prompt injection, the risk is mitigated by the skill's structured output requirements and specific instructions to verify data sources and flag preprints as non-peer-reviewed. The agent's capabilities are primarily restricted to information retrieval and report generation.\n- [EXTERNAL_DOWNLOADS]: The instructions require the installation of the esm package directly from a specific commit on the official EvolutionaryScale GitHub repository to enable mechanism-of-effect analysis using Sparse Autoencoders. This represents a standard dependency for accessing specific research features not yet available in stable release channels.

tooluniverse-variant-interpretation