arkcli-models

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data returned by the arkcli tool, such as model names, descriptions, and metadata. This introduces a surface for indirect prompt injection if the remote registry contains adversarial content designed to influence the agent's behavior.
  • Ingestion points: Results from arkcli models list and arkcli models search in SKILL.md and referenced files.
  • Boundary markers: None explicitly mentioned in the scripts for model data processing.
  • Capability inventory: Command execution via arkcli and local script execution using python3.
  • Sanitization: The provided processing scripts in references/arkcli-models-list.md do not sanitize string fields before outputting them to the agent context.
  • [DYNAMIC_EXECUTION]: The skill provides a Python script template in references/arkcli-models-list.md for the agent to execute locally. This script is intended to filter and process JSON output where the CLI tool lacks server-side filtering capabilities.
  • Evidence: A bash heredoc pattern in references/arkcli-models-list.md that pipes a multi-line Python script to python3 - to handle date parsing and item filtering.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 08:05 AM
Security Audit — agent-trust-hub — arkcli-models