cancel-reason-classifier

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-generated free-text reasons, which introduces a potential surface for indirect prompt injection if the agent follows the suggested LLM fallback workflow.
  • Ingestion points: Free-text data enters the system through the cancel_reason_raw field in external cohort JSON files.
  • Boundary markers: The workflow description does not specify the use of delimiters or 'ignore' instructions when passing residues to the LLM for classification.
  • Capability inventory: The skill performs file system operations (read/write) and executes a local Python script for keyword matching.
  • Sanitization: No sanitization or filtering is applied to the raw input text beyond basic lowercasing.
  • [SAFE]: The classification logic is primarily deterministic and utilizes yaml.safe_load() for configuration parsing, effectively mitigating risks associated with untrusted YAML data.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 05:18 AM
Security Audit — agent-trust-hub — cancel-reason-classifier