npm security reports, with evidence.

This is an automatic install-time mutation of a broad, foreign AI-agent control surface that creates persistent command execution. The conditional check for existing binaries does not pro...

The hard-coded private endpoint, automatic importer-side activation, and posting of serialized agent inputs and outputs establish concrete data exfiltration behavior. The lack of an npm l...

The hard-coded credential forwarding is not necessary for local package operation and directs a caller secret to an unrelated gateway. Absence of lifecycle hooks reduces automatic-install...

The hard-coded trace collector and automatic runtime configuration create a concrete, unconsented data-exfiltration path. No lifecycle hook is needed for the harmful behavior to occur dur...

The package contains a concrete credential-exfiltration path activated during normal SDK transaction and quote operations. Lack of install hooks does not mitigate remote recovery of consu...

The package has an automatic lifecycle trigger and directly exfiltrates identifying host data to a fixed external endpoint. This is concrete malicious install-hook abuse, not package-alig...

The automatic postinstall hook performs a concrete, broad process-control action against running agent bridges. It does not need credential theft to create an unconsented install-time con...

This is a concrete remote-code-execution chain: remotely mutable web content reaches an unguarded bridge that executes host shell commands. The absence of an npm install hook does not mit...

The automatic lifecycle trigger, deliberate obfuscation, broad agent-skill replacement, and telemetry form a concrete malicious install-hook chain. The commerce functionality does not jus...

The package implements a concrete foreign-credential collection, remote bearer-token transmission, and cross-application credential persistence chain. No install hook is present, but the...

The package automatically persists a launcher in Claude Desktop configuration during postinstall. This meets the policy for malicious install-hook abuse regardless of the Warp-branded API...

This is an unconsented postinstall mutation of a broad AI-agent control surface. The source makes the persistence automatic rather than requiring an explicit setup command.

This is a concrete unconsented postinstall mutation of a broad AI-agent control surface. The package suppresses installation failures and deliberately bypasses interactive consent.

Source directly connects token-bearing command options to a default external telemetry capture path. The absence of npm lifecycle hooks does not mitigate this runtime credential disclosure.

The package contains automatic, hard-coded exfiltration of user-controlled agent tool data to an unrelated remote logging service. Limited pattern redaction does not remove the concrete p...

This is an unconsented install-time mutation of a broad AI-agent control surface, which meets the blocking policy. The explicit feedback endpoint is not the basis for the verdict.

This is a concrete automatic postinstall chain that broadly mutates a foreign AI-agent control surface and adds persistent package-controlled hooks and instructions. The absence of proven...

This is an unconsented postinstall mutation of a broad AI-agent control surface. The lifecycle path is explicit and directly creates global hooks and global instruction files.

The fixed xAI TTS request forwards the active runtime credential without ensuring it is an xAI credential. This is concrete credential exfiltration reachable from the package CLI.

This is a concrete automatic postinstall control-surface mutation, which meets the blocking policy. No network exfiltration was needed to establish the install-hook abuse.