acestep-lyrics-transcription
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes audio transcription data from OpenAI and ElevenLabs APIs, which the agent is then instructed to read and correct manually. This creates an ingestion point for untrusted content.
- Ingestion points: The agent reads transcribed lyrics from .lrc, .srt, or .json files generated in the acestep_output directory.
- Capability inventory: The agent has access to Read, Write, and Bash tools, which could be targeted if the transcribed text contained malicious instructions that the agent mistakenly followed during the correction phase.
- Boundary markers: While the skill provides clear instructions on how to perform corrections, it does not include explicit delimiters or 'ignore' instructions for the content of the transcribed files themselves.
- Sanitization: The acestep-lyrics-transcription.sh script performs no validation or sanitization of the text returned by the transcription APIs before the agent processes it.
- [COMMAND_EXECUTION]: The bash script uses standard utilities including curl for API communication with well-known services (OpenAI and ElevenLabs), jq for JSON processing, and python for data conversion.
- [DYNAMIC_EXECUTION]: The skill executes embedded Python code via python -c to handle word-level timestamp processing and format conversion. This logic is static within the bash script and does not involve executing dynamically generated code from external inputs.
Audit Metadata