[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Context & Harness Attacks · markdown

official (OSV malicious-package reports MAL-2026-20187/-20195/-20196 and GitHub malware advisories, Nov 18-19, 2026) · published here

Five more malicious npm packages pose as AI SDKs and MCP servers, including a Markdown MCP server that grabs Alibaba Cloud instance metadata on load

  • Search lockfiles, global installs, the npx cache and MCP client configs (Claude Desktop, Cursor, VS Code, Windsurf) for the five names.
  • If one ran: treat the host as compromised and rotate cloud, SSH, Git and AI-provider keys; on Alibaba Cloud VMs check for output/field_check.txt and RAM role credential use.
  • Prevent: pin MCP servers to vetted versions instead of npx -y latest.

Sources: OSV MAL-2026-20196: Malicious code in character-ai-sdk (npm), via Amazon Inspector (Nov 19, 2026) · OSV MAL-2026-20187: Malicious code in @fdcasj_dfew/md-doc-mcp (npm) (Nov 18, 2026) · GitHub malware advisory GHSA-2r45-7hg7-w99r: @mibbp/agent-observer (Nov 19, 2026) · GitHub malware advisory GHSA-xvx7-m3mm-3jpw: mrnima-openai-chatgpt-bot (Nov 19, 2026)