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

← Context & Harness Attacks · markdown

security-vendor research (Socket Threat Research, Nov 6); no confirmed victims; gem removed by RubyGems after ~5 hours

Malicious Ruby gem carried a Shai-Hulud-family worm that steals AI coding tool credentials and plants MCP servers, SessionStart hooks and AGENTS.md in repos

  • Exposure window: about five hours on Nov 1 before RubyGems removed both versions. A PyPI sibling, pystr-codec 0.1.0 to 0.1.2, uses the same paleasbest[.]top endpoint.
  • Why it matters for agents: opening an infected repository in an AI coding tool or VS Code can re-run the worm, depending on that tool's workspace trust and approval settings; the generated AGENTS.md tells agents to run a setup script first.
  • Order matters if infected: isolate the host and remove net-health-checker and pkg-runtime before revoking the GitHub token, or the wiper fires.

Sources: Socket: Shai-Hulud descendant surfaces in malicious Ruby gem (Nov 6)