jcodemunch-mcp-code-retrieval

Warn

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to install the jcodemunch-mcp package from the public PyPI registry using pip or uvx. These are unverified external dependencies that are not from a trusted vendor.
  • [REMOTE_CODE_EXECUTION]: The installation methods provided, specifically uvx jcodemunch-mcp, involve the direct execution of a remotely hosted package. Additionally, the jcm CLI utility is used to download and run setup scripts.
  • [COMMAND_EXECUTION]: The skill employs a CLI tool (jcm) that executes shell commands to modify system-level MCP configuration files for various clients like VS Code, Cursor, Windsurf, and Claude Code.
  • [INDIRECT_PROMPT_INJECTION]: The core functionality involves ingesting and parsing untrusted data from local codebases to serve as agent context.
  • Ingestion points: Local source code files processed via index_repository, find_symbols, and get_symbol_content.
  • Boundary markers: Absent; the skill does not specify the use of delimiters or 'ignore' instructions when interpolating retrieved code into the agent prompt.
  • Capability inventory: The skill provides high-privilege context retrieval tools including get_symbol_content, assemble_task_context, and get_repo_map, which can expose the agent to a large volume of untrusted data.
  • Sanitization: Absent; the tool retrieves and returns raw source code implementations based on tree-sitter AST parsing without filtering for malicious instruction patterns.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 16, 2026, 10:44 PM
Security Audit — agent-trust-hub — jcodemunch-mcp-code-retrieval