primekg

Pass

Audited by Gen Agent Trust Hub on May 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates indirect prompt injection by ingesting data from an external knowledge graph source (kg.csv) and interpolating it into the agent's context without sanitization or boundary markers.
  • Ingestion points: scripts/query_primekg.py (via _load_kg and pd.read_csv).
  • Boundary markers: Absent; data is returned directly to the agent as dictionary records.
  • Capability inventory: Reads local files using pandas.read_csv and performs string filtering.
  • Sanitization: No escaping or validation is performed on the content retrieved from the CSV before it is passed to the agent.
  • [COMMAND_EXECUTION]: All data processing is performed using the pandas library. No shell commands or subprocess calls were detected.
  • [DATA_EXPOSURE]: The skill accesses a hardcoded local file path (/mnt/c/Users/eamon/Documents/Data/PrimeKG/kg.csv). While this reveals internal directory structures, the access is confined to the intended scientific dataset and does not target sensitive system or configuration files.
Audit Metadata
Risk Level
SAFE
Analyzed
May 29, 2026, 02:39 PM
Security Audit — agent-trust-hub — primekg