Skills
Navigate
skills/jeremylongshore/tons-of-skills-marketplace/deepgram-common-errors/Gen Agent Trust Hub

deepgram-common-errors

Warn

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The splitAudio function in references/implementation.md uses child_process.exec with direct string interpolation of the inputPath variable. \n
  • Evidence: await execAsync(ffmpeg -i ${inputPath} -f segment -segment_time ${chunkDuration} -c copy ${outputPattern}) \n
  • Risk: If an attacker can control the filename of an audio file processed by this skill, they could execute arbitrary shell commands. \n- [INDIRECT_PROMPT_INJECTION]: The skill processes external audio files and application logs which are potential vectors for indirect prompt injection. \n
  • Ingestion points: Audio files read via fs.readFileSync and log files referenced in instructions (SKILL.md, references/implementation.md). \n
  • Boundary markers: None identified in code snippets to distinguish between data and instructions. \n
  • Capability inventory: Shell command execution via exec, file system reads, and network requests via curl (SKILL.md, references/implementation.md). \n
  • Sanitization: No sanitization or validation of input paths or file content is present before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 9, 2026, 03:43 AM