arkcli-custommodel

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from the custom model repository (such as model names and descriptions) through arkcli models custommodel list and get commands. \n
  • Ingestion points: Untrusted metadata enters the agent context through CLI outputs from model listing and detail retrieval (SKILL.md, arkcli-custommodel-get.md). \n
  • Boundary markers: No explicit delimiters or boundary instructions are defined in the skill files for separating CLI output from agent instructions. \n
  • Capability inventory: The skill allows for the creation, modification, and deletion of resources via upload, update, delete, and quantize commands (SKILL.md, arkcli-custommodel-delete.md). \n
  • Sanitization: The skill instructions enforce parameter validation for the sort-order flag, preventing invalid input from being passed to the tool. \n- [COMMAND_EXECUTION]: The skill performs administrative tasks by executing shell commands via the arkcli binary. \n
  • Evidence: Commands for managing model lifecycles (upload, get, list, update, delete, quantize) are central to the skill. The instructions mitigate risk by mandating user confirmation before write operations and providing dry-run options for previews.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 04:04 PM
Security Audit — agent-trust-hub — arkcli-custommodel