AI-agent control-surface risk in npm packages.
These packages silently write AI-agent control files during install: agent instructions, MCP server entries, hooks or permission settings for tools such as Claude, Codex or Cursor. This is a dangerous-capability policy finding, not necessarily malware. LPM CLI warns by default and blocks when Firewall is configured to.
Packages — page 13
579 packages with LPM's own review, newest scan first. Showing 577–579.
Direct source inspection confirms a concrete AI-agent control-surface mutation that exfiltrates agent transcripts to a hardcoded remote endpoint without clear user-facing disclosure in RE...
The package performs unconsented lifecycle writes to OpenCode agent/control files with shell/edit-capable agents, matching the firewall's malicious AI-agent control hijack boundary. Other...
Direct source inspection confirms unconsented lifecycle writes into AI-agent control-surface directories from remote content, which meets the firewall blocking criterion for AI-agent cont...