arkcli-understand
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill requires and executes the arkcli binary to perform multimodal analysis, resource resolution, and authentication checks.
- [DATA_EXFILTRATION]: Local files provided via the --input flag, including images, video, and audio, are transmitted to the volcengine infrastructure for processing. Audio data is specifically inlined as base64 data URLs for client-side processing before transmission.
- [PROMPT_INJECTION]: The skill provides --system-prompt-override and --system-prompt-append flags, allowing for the modification of internal system instructions used by the analysis engine.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from untrusted external data. 1. Ingestion points: Local and remote files supplied via the --input flag in SKILL.md. 2. Boundary markers: The instructions differentiate between recipe instructions and user prompts but lack robust isolation for file content. 3. Capability inventory: The skill uses arkcli to read local files and perform network requests to the arkruntime API. 4. Sanitization: No specific filtering or sanitization of external file content is implemented.
- [EXTERNAL_DOWNLOADS]: The skill supports fetching and processing remote files from https and tos URLs.
Audit Metadata