alphafold-database
Warn
Audited by Gen Agent Trust Hub on Apr 29, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs several well-known scientific and cloud-related Python packages, including biopython, requests, google-cloud-bigquery, and gsutil, from standard registries.
- [DATA_EXFILTRATION]: The skill performs network requests to established scientific and technology domains (alphafold.ebi.ac.uk, uniprot.org, and google.cloud) to retrieve protein metadata and structure files.
- [COMMAND_EXECUTION]: The skill includes a 'download_proteome' function in SKILL.md that executes shell commands via 'subprocess.run(..., shell=True)'. The command incorporates a 'taxonomy_id' variable directly into the shell string without sanitization, which presents a command injection risk if the input is sourced from an untrusted user or external data.
- [PROMPT_INJECTION]: The skill functions as a surface for indirect prompt injection by ingesting data from external scientific APIs and parsing protein structure files.
- Ingestion points: API responses from AlphaFold and UniProt endpoints; local .cif and .json data files.
- Boundary markers: None detected in the implementation.
- Capability inventory: Execution of shell commands via subprocess, network access, and file system writes.
- Sanitization: No input validation or filtering is present for the data processed by the skill.
Audit Metadata