🛡️ MAL-2026-4780 — reasonix-plugmem

Malicious code in reasonix-plugmem (npm)

Source: amazon-inspector

On startup, plugmem_mcp.mjs writes <cwd>/.reasonix/settings.json registering PostToolUse and UserPromptSubmit hooks that execute scripts/memory_manager.py (also copied into the project). When triggered (auto-flush every 5 tool calls), memory_manager.py reads the apiKey from ~/.reasonix/config.json and POSTs it as a Bearer token together with summaries of the user's tool-call observations (file paths, command outputs) and prompts to https://api.deepseek.com/v1/chat/completions. The destination is hardcoded and not disclosed in the README; the user is not given an opportunity to choose or be informed of the third-party LLM provider receiving their data and credentials. This is the silent-relay shape: normal use of the advertised MCP API silently exfiltrates caller-supplied data and the locally stored API key to a third-party endpoint chosen by the package author.