find-music
Pass
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The
scripts/gen_music.pyscript downloads audio files from external URLs returned by the Fal CassetteAI API (requests.get(url).content). While the service is a specific music generation provider, it constitutes downloading content from a remote server. - [DATA_EXFILTRATION]: The script
scripts/gen_music.pyreads sensitive credentials from a.envfile (FAL_API_KEY) and sets them as environment variables (FAL_KEY) before making network requests to an external API. While this is standard for API-driven skills, it involves handling sensitive API keys. - [COMMAND_EXECUTION]: The script
scripts/beat_detect.pyandreferences/example-music.pyperform file system operations, including writing audio files and JSON manifests to disk.references/example-music.pyalso writes to an absolute file path on a specific user's system (/Users/shivsakhuja/...), which indicates hardcoded environment assumptions. - [INDIRECT_PROMPT_INJECTION]: The
scripts/gen_music.pyscript uses aVIBESdictionary to construct prompts for an external AI music generator. User-provided briefs (documented inSKILL.mdandtests/sample-input.md) are the primary input for determining which prompt to use, creating a surface where malicious input could influence the generated music prompts. Severity is LOW as it follows the skill's intended purpose. - Ingestion points: User briefs and source asset paths defined in
SKILL.mdandtests/sample-input.md. - Boundary markers: None detected; the skill relies on confirmation of inputs but lacks explicit delimiters for untrusted data.
- Capability inventory: File writes, network requests (Fal API), and audio processing (librosa).
- Sanitization: No evidence of input sanitization or filtering of the user brief before it influences the workflow.
Audit Metadata