The Agent Skills Directory

External Data Retrieval: The skill fetches structural and sequence data from established scientific repositories (PDB and UniProt). This is a standard part of the scientific workflow and targets reputable, well-known domains.
Command Orchestration: The skill automates the execution of bioinformatics tools and the generation of cluster job manifests (Kubernetes and Slurm). These operations are performed within a user-controlled environment and are transparently logged in a dedicated directory.
Generated scripts and logs are stored in the jobs/ folder for review.
The skill supports least-privilege deployment models for cluster environments.
Data Processing Surface: The skill processes external protein data which could theoretically be used for indirect prompt injection. However, the skill explicitly includes internal checks to verify that the agent correctly distinguishes between instructional content and data.
The ground-truth.yaml file includes specific test cases for the agent to identify and ignore potential instructions embedded in data files.
Secure Scripting: The provided utility script for sequence profiling is implemented using standard libraries and focuses on deterministic calculations.
scripts/sequence_profile.py includes input validation to ensure sequences consist of standard amino acids and features a self-test routine to verify operational integrity.
Transparency and Governance: The skill documentation includes sections on data handling, declared use cases, and access levels, providing clear guidance on monitoring and scope compliance.

binder-gauntlet