cosmic-database

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection] (LOW): The skill processes genomic data files from external sources, which provides a surface for indirect prompt injection attacks.\n
  • Ingestion points: Untrusted data is downloaded via scripts/download_cosmic.py and read into the agent context using pandas.read_csv and pysam.VariantFile as shown in SKILL.md.\n
  • Boundary markers: Absent. The documentation and scripts do not implement delimiters or instructions to disregard potential commands embedded within the genomic data files.\n
  • Capability inventory: The skill possesses file-writing capabilities in scripts/download_cosmic.py (line 103). It does not include high-risk capabilities like subprocess execution or dynamic code evaluation (eval/exec).\n
  • Sanitization: Absent. No sanitization or validation logic is applied to the contents of the downloaded data files before they are processed by the agent.\n- [SAFE] (SAFE): No malicious code or exfiltration patterns were detected. All network communication is restricted to the legitimate COSMIC database domain at cancer.sanger.ac.uk. Credential handling is standard for a command-line research utility.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:24 PM