The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The usage-guide.md instructs users to install software directly from unverified GitHub repositories, such as pip install git+https://github.com/tkzeng/Pangolin.git and references github.com/kidsneuro-lab/SpliceVault. These sources are not within the recognized trusted vendor list and pose a supply chain risk.
[COMMAND_EXECUTION]: The script examples/spliceai_clingen_classify.py uses subprocess.run() to execute the spliceai command-line interface. While this is necessary for the skill's primary function, it grants the agent the ability to execute system commands using local file paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes VCF (Variant Call Format) files, which are external, potentially untrusted inputs. The agent reads and interprets data from these files (specifically the INFO fields) to perform ClinGen SVI classifications, creating an attack surface for indirect prompt injection.
Ingestion points: The parse_spliceai_vcf function in examples/spliceai_clingen_classify.py reads and parses the INFO column of VCF files.
Boundary markers: None identified; the parsed data is interpolated into the agent's context without explicit isolation or instructions to ignore embedded commands.
Capability inventory: The skill possesses file system access (read/write) via the pathlib and open() functions, as well as command execution capabilities via subprocess.run().
Sanitization: The script employs basic regex for extraction but lacks sanitization or validation to prevent the agent from acting on malicious instructions hidden within the genomic data fields.

bio-splice-variant-prediction