musia-lalachan-mv-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell scripts (
scripts/mux_musia_audio.sh,scripts/wait_downloaded_mp4.sh) to perform media operations. These scripts execute system binaries includingffmpegfor audio/video muxing andfind/statfor monitoring the~/Downloadsdirectory. The scripts handle arguments with standard shell quoting to mitigate simple command injection. - [INDIRECT_PROMPT_INJECTION]: The workflow involves generating prompts for the Xiaoyunque AI agent based on data ingested from local creative project files, such as
STORY.mdandASSET_LIST.mdlocated in theSKILL.mddefined directories. - Ingestion points: Project metadata and content located in
$MUSIA_ROOT/data/creative_projects/are used to build storyboards and prompts. - Boundary markers: The skill lacks explicit instructions for the agent to sanitize or escape song titles or story content before they are interpolated into the generated Xiaoyunque prompts.
- Capability inventory: The skill possesses the ability to execute shell commands via
ffmpegandffprobe, and perform file-write operations within project directories. - Sanitization: While the skill provides explicit instructions to avoid including local filesystem paths in prompts (preferring asset labels like
图1), it does not specify sanitization for user-generated text fields that form the core of the story and visual instructions.
Audit Metadata