clean-transcription
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill implements automated text processing workflows that execute shell commands such as awk and sed to perform regex-based substitutions and maintain the structural integrity of SRT and TXT files.
- [PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill ingests untrusted external data from source transcripts (.srt, .txt) and glossary files (glossary.json). This content is interpolated into prompts during the optional LLM polish phase, potentially allowing malicious content within the transcript to influence agent behavior.
- Ingestion points: Source transcript files (.srt, .txt) and project-level glossary.json files.
- Boundary markers: The skill includes logic to validate that cue counts and timestamps are preserved between input and output.
- Capability inventory: Shell command execution (awk, sed, grep, etc.) and file system modification (Write, Edit).
- Sanitization: Employs post-processing validation of file structure (cue count checks).
Audit Metadata