musia-fun-website-item

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local shell commands to audit, validate, and curate media files. Evidence: Instructions use node bin/musia.js, conda run -n musia python scripts/run_pipeline.py, and npm run website:validate to perform automation tasks within the local repository.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted audio-to-text transcriptions and reference lyrics that are used to update catalog entries and website metadata. 1. Ingestion points: ASR/STT model results and reference lyric files described in the workflow section. 2. Boundary markers: Absent; transcriptions and lyric content are interpolated into JSON manifests without explicit delimiters. 3. Capability inventory: The skill modifies catalog.json and manifest.json files and executes processing scripts with file-system write access. 4. Sanitization: No explicit validation or filtering of transcript content is performed before it is added to the catalog.
  • [DATA_EXFILTRATION]: The skill mandates copying generated video assets to an external synchronization folder for cloud storage. Evidence: The workflow directs the agent to copy recordings to /home/lachlan/Nutstore Files/Projects/Musia/ for synchronization.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:24 PM