knowledge-refiner

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process untrusted unstructured content from the work/ and knowledge/ directories. It lacks explicit boundary markers or sanitization protocols for this data. Since the skill has the capability to write to specified output paths (out argument) and influence the agent's internal knowledge base, a malicious payload within a processed note could redirect the agent to leak information or corrupt the knowledge repository.
  • [Data Exposure & Exfiltration] (MEDIUM): The 'Knowledge Protocol' section indicates the skill interacts with 'Confidential (Company/Client)' and 'Personal' knowledge tiers, including secrets. While no hardcoded credentials are found, the functionality that merges these tiers into structured files creates a risk where sensitive data could be consolidated and then exfiltrated if the output path or refinement logic is manipulated.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 12:58 PM