alfworld-goal-interpreter

Pass

Audited by Gen Agent Trust Hub on Jul 7, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions are limited to task parsing and environment interaction logic for the ALFWorld simulator. No malicious commands, data exfiltration, or obfuscation were found.
  • [COMMAND_EXECUTION]: The skill references a local Python script 'parse_goal.py' for goal parsing. This script is a vendor resource from 'zjunlp' and is part of the standard architectural pattern for this skill's task decomposition.
  • [PROMPT_INJECTION]: The skill processes untrusted natural language task strings as input, creating a potential surface for indirect prompt injection. Ingestion points: raw task strings in SKILL.md. Boundary markers: none. Capability inventory: uses parse_goal.py for processing. Sanitization: none. This is an inherent risk of the skill's primary purpose and is assessed as safe in this context.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 7, 2026, 01:05 PM
Security Audit — agent-trust-hub — alfworld-goal-interpreter