jimeng-cli-text2video
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and update the dreamina CLI tool from the official Jimeng domain (jimeng.jianying.com), which is the legitimate host for this service.
- [REMOTE_CODE_EXECUTION]: Includes a command to pipe a remote script to bash (
curl -fsSL https://jimeng.jianying.com/cli | bash) for tool installation. This originates from an official domain of a well-known technology provider and follows standard deployment practices for CLI utilities. - [COMMAND_EXECUTION]: Utilizes various dreamina CLI commands for text-to-video generation, task polling, and account management. The execution flow is clearly defined with logical separation between submission and result retrieval.
- [DATA_EXFILTRATION]: Mentions local configuration paths (e.g.,
~/.dreamina_cli/credential.json) solely for existence checks and troubleshooting. No instructions were found that read or transmit sensitive file contents to external destinations. - [PROMPT_INJECTION]: The skill includes explicit instructions for the agent to seek user approval before executing commands. It further mitigates risks by requiring prompt preparation through a dedicated upstream skill methodology.
- [INDIRECT_PROMPT_INJECTION]: An attack surface exists where user-provided prompts are interpolated into shell commands. This is addressed by the following evidence chain:
- Ingestion points: User-supplied or upstream-crafted prompt strings in SKILL.md and multiple examples.
- Boundary markers: Mandatory human-in-the-loop review and requirement for upstream skill validation before execution.
- Capability inventory: Shell execution capability via the dreamina CLI tool.
- Sanitization: Relies on mandatory user approval and pre-validation by a specialized prompt-crafting skill.
Audit Metadata