jlceda-mcp-automation

Warn

Audited by Gen Agent Trust Hub on Oct 20, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to clone repositories from unverified GitHub accounts, specifically hyl64/jlcmcp. It also recommends repositories from sengbin and Spectoda for various PCB automation tasks.
  • [REMOTE_CODE_EXECUTION]: The automation process involves running npm run build on the contents of cloned external repositories to generate executable bridge components, effectively executing code from unverified sources.
  • [DATA_EXFILTRATION]: The skill requires access to sensitive local file paths, including the Chrome browser profile directory at $HOME/.cache/xyq-chrome (which contains cookies, history, and session data) and software activation files (~/Downloads/lceda-pro-activation.txt). While instructions exist to avoid printing this data, the access itself constitutes a data exposure surface.
  • [COMMAND_EXECUTION]: The skill executes multiple local shell and Python scripts located in agentic_tools/jlceda_mcp_agent/scripts/ to manage installation, launching, and activation of the EDA software.
  • [INDIRECT_PROMPT_INJECTION]: This skill serves as a vulnerability surface by ingesting untrusted data from external repositories and local files while possessing broad execution capabilities (subprocesses, file writes, and network gateway access).
  • Ingestion points: Third-party GitHub repositories and local activation files.
  • Boundary markers: Includes instructions to not print or commit sensitive contents, though these are natural language constraints rather than technical sanitization.
  • Capability inventory: Subprocess execution (node, python, bash), file system writes (writing build outputs and tools lists), and network operations (WebSocket bridge on port 18800).
  • Sanitization: None specified for the data read from external sources or activation files before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 20, 2026, 06:39 AM