The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and documentation describe the installation of the gtars library and gtars-cli via PyPI and crates.io using pinned versions (gtars==0.9.2, gtars-cli==0.9.0). These are well-known package registries.- [EXTERNAL_DOWNLOADS]: The Tokenizer.from_pretrained and RefgetStore.open_remote methods can download genomic artifacts and metadata from Hugging Face Hub and other remote services. The skill mitigates risks by recommending local snapshots, revision pinning, and SHA-256 verification.- [COMMAND_EXECUTION]: The skill involves executing native-code extensions and CLI tools. It provides an artifact_inspector.py script to hash and classify artifacts without loading or executing them, encouraging a 'verify-before-install' security posture.- [DATA_EXFILTRATION]: Certain library functions, such as RegionSet(path) with its URL fallback and RefgetStore.open_remote, possess network capabilities that could inadvertently transmit sensitive genomic metadata or coordinates. The skill documentation explicitly warns about these side effects and provides redaction-enabled local reporting scripts.- [INDIRECT_PROMPT_INJECTION]: As the skill ingests untrusted genomic data files (BED, FASTA, fragments), it is vulnerable to indirect prompt injection if the data contains malicious patterns targeting the processing logic.
Ingestion points: Data ingestion occurs through file-backed RegionSet constructors, load_fasta sequences, and barcode tokenization routines.
Boundary markers: The skill mandates the use of pre-execution validators (bed_validator.py, execution_plan.py) and coordinate contracts.
Capability inventory: The gtars native extension and CLI support complex genomic algebra, signal track generation, and network-enabled artifact retrieval.
Sanitization: The scripts/_common.py module implements strict validation, including NUL byte checks, path traversal rejection, and genomic coordinate bounds enforcement.

gtars