yuque-document-management

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • Indirect Prompt Injection (HIGH): The yuque-document-management skill (SKILL.md) processes external content via mcp_YuQueMCP_get_document. This creates a vulnerability where malicious document content could hijack the agent's logic to execute destructive tools. \n- Ingestion points: mcp_YuQueMCP_get_document in SKILL.md. \n- Boundary markers: None identified. \n- Capability inventory: delete_document, update_document, move_document in SKILL.md. \n- Sanitization: None identified.\n- Command Execution (MEDIUM): The web-artifacts-builder and webapp-testing components execute shell commands and manage server lifecycles through scripts like scripts/init-artifact.sh and scripts/with_server.py. Specifically, with_server.py executes user-provided strings as shell commands to start servers.\n- External Downloads (MEDIUM): The slack-gif-creator and web-artifacts-builder skills trigger runtime installation of multiple Python and Node.js packages (e.g., pillow, imageio, parcel, html-inline) which could lead to supply chain attacks if dependencies are compromised.\n- Dynamic Execution (MEDIUM): web-artifacts-builder utilizes Parcel to bundle code into self-contained HTML artifacts, involving runtime asset processing and inlining.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 11:38 PM