curate-livestream-transcripts

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to fetch and install a skill builder utility from a GitHub repository associated with the author (github.com/peter-tu-zynkr/zynkr-skill-builder).
  • [REMOTE_CODE_EXECUTION]: The provided setup command (npx skills add ...) downloads and executes code from a remote source during the skill installation phase.
  • [COMMAND_EXECUTION]: The skill uses specific Model Context Protocol (MCP) tools to perform automated operations on Google Drive, including listing folder contents, reading files, and creating new documents.
  • [PROMPT_INJECTION]: The skill processes transcript data retrieved from external Google Docs to generate structured summaries. This creates a surface for indirect prompt injection where content within the transcripts could attempt to influence the agent's summarization behavior.
  • Ingestion points: The skill reads raw 逐字稿 (transcript) text from source Docs found in specified upstream Drive folders (SKILL.md, Step 6).
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' prompts to isolate the processed transcript data from the system instructions.
  • Capability inventory: The skill is capable of creating and modifying Google Docs (create_doc, create_file, update_drive_file) based on the summarized content.
  • Sanitization: There is no evidence of transcript content sanitization or validation before it is interpolated into the summarization prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 01:35 AM
Security Audit — agent-trust-hub — curate-livestream-transcripts