The Agent Skills Directory

[SAFE]: The skill is developed with an exceptionally high degree of security awareness, providing specialized scripts to validate all inputs before processing. These scripts implement strict safety contracts that reject path traversal, symlinks, and NUL bytes while enforcing resource bounds on file size and record counts.
[EXTERNAL_DOWNLOADS]: The skill references official package repositories including PyPI, crates.io, and the project's official GitHub organization (databio). It also details optional network operations for fetching pretrained tokenizers from Hugging Face and genomic sequence collections from BEDbase APIs. All external references are to well-known or project-official services and are documented with clear integrity verification and approval gates.
[INDIRECT_PROMPT_INJECTION]: The skill processes external genomic data which could theoretically contain malicious instructions. This risk is effectively mitigated by a mandatory validation workflow using bundled scripts (e.g., bed_validator.py and tokenizer_manifest.py) that ensure data strictly adheres to genomic schemas and does not exceed memory or coordinate limits.

gtars