The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and reference files encourage using the subprocess module to execute external bioinformatics tools such as blastn, makeblastdb, clustalo, muscle, and dssp. These are standard tools in bioinformatics, but they represent a command execution capability via the skill's Bash tool access.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted biological data from external files and database records, creating an attack surface for indirect prompt injection.
Ingestion points: Biological files (FASTA, GenBank, PDB, etc.) are read from the local file system using Bio.SeqIO, Bio.AlignIO, Bio.PDB, and Bio.Phylo modules. Records are also fetched dynamically from NCBI databases.
Boundary markers: The instructions do not contain specific delimiters or warnings for the agent to ignore instructions that might be embedded in record descriptions, metadata, or sequences.
Capability inventory: The agent has access to the shell via the Bash tool and is explicitly taught to use subprocess.run to execute binaries.
Sanitization: While data is parsed into structured objects, there is no evidence of filtering or sanitizing natural language metadata that might be presented to the agent's context.
[EXTERNAL_DOWNLOADS]: The skill downloads biological sequences, structure files, and publication metadata from well-known and trusted scientific services including the National Center for Biotechnology Information (NCBI) via Entrez, the Protein Data Bank (PDB), and PubChem. These operations are neutral and intended for the skill's primary purpose.

alterlab-biopython