music-composer

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script compose_music.py executes the ffmpeg binary using subprocess.run to transcode synthesized audio into MP3 format. The implementation is secure, utilizing a list of arguments rather than a raw shell string to prevent command injection, and it resolves the binary path using the system's PATH via shutil.which.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided storyline.json files, which presents a surface for indirect prompt injection. 1. Ingestion points: Data is loaded in compose_music.py using json.loads. 2. Boundary markers: The instructions do not include delimiters or warnings to ignore instructions within the JSON content. 3. Capability inventory: The skill can write multiple audio and metadata files and execute the ffmpeg command. 4. Sanitization: The script validates data types but does not sanitize free-text metadata fields like title or comment that are output to logs.\n- [SAFE]: The skill operates entirely locally and does not perform unauthorized network requests or access sensitive files. All external links, such as the repository at github.com/ellmos-ai/ai-media-editor and references to tools like FluidSynth, are to official and well-known resources consistent with the skill's purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 05:44 AM