skills/jimmc414/kosmos/hypogenic/Gen Agent Trust Hub

hypogenic

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • Prompt Injection (LOW): The configuration defines prompt templates that interpolate untrusted data from JSON files and PDF documents. This represents a surface for indirect prompt injection.\n
  • Ingestion points: data/train.json, data/val.json, data/test.json, and files in the papers/ directory.\n
  • Boundary markers: None. Data is directly inserted into templates using placeholders like {data_samples} and {sample_text} without delimiters or instructions to ignore embedded commands.\n
  • Capability inventory: The skill uses LLMs for hypothesis generation, inference (classification), and adaptive refinement.\n
  • Sanitization: No sanitization or escaping logic is present in the configuration templates to handle potentially malicious content within the dataset or literature.\n- Credentials (SAFE): API keys are referenced via environment variables (OPENAI_API_KEY) rather than being hardcoded.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:22 PM