The Agent Skills Directory

[COMMAND_EXECUTION]: In SKILL.md, the download_proteome function uses subprocess.run(cmd, shell=True, check=True) where the command string is constructed using f-strings from variables taxonomy_id and output_dir. This pattern is vulnerable to command injection if an agent populates these parameters with unsanitized input from a user prompt.
[EXTERNAL_DOWNLOADS]: The skill fetches configuration and data from alphafold.ebi.ac.uk and uniprot.org, and accesses public datasets on Google Cloud Storage (gs://public-datasets-deepmind-alphafold-v4). These are well-known scientific and cloud services, and the references are consistent with the skill's purpose.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes external data from scientific APIs without explicit boundary markers.
Ingestion points: Data is ingested from REST API responses (AlphaFold/UniProt) and BigQuery metadata in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used in the provided Python examples.
Capability inventory: The skill can execute shell commands (subprocess.run), perform network requests (requests, urllib), and write to the file system.
Sanitization: The provided Python examples do not demonstrate validation or sanitization of external data before it is used in command execution or further processing.

alphafold-database