arkcli-chat
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data via prompts and file attachments, creating a surface for potential instruction injection.\n
- Ingestion points: Positional
<prompt>arguments and files provided via the--inputflag enter the execution environment as described inSKILL.mdandreferences/arkcli-chat.md.\n - Boundary markers: The skill instructions emphasize the use of
--instructionsfor system prompts and the--text-strictflag for enforcing output structure via JSON schemas.\n - Capability inventory: The skill utilizes shell command execution for the
arkclitool and performs file operations, including reading and writing JSON response data usingjqand shell redirection, as detailed inreferences/text-format.md.\n - Sanitization: The skill provides client-side validation logic for model output and includes explicit warnings to prevent the leaking of API keys during execution.
Audit Metadata