npm detections from LPM Firewall.

The manifest activates the bootstrap during npm postinstall, and source shows it copying package-controlled hook scripts and helpers into an existing consumer .claude directory. That unco...

The automatic postinstall hook writes user-level Claude telemetry settings that enable prompt and assistant-response logging to the package’s configured collector. This is unconsented mut...

The entrypoint automatically disrupts selected visitors’ pages after a delay, including by blocking pointer interaction and playing external audio. This is active targeted browser disrupt...

The bundle's targeted page disruption and unsolicited political audio are active protestware behavior, not ordinary import/export functionality. The manifest exposes this bundle as the pa...

The entrypoint contains targeted, delayed browser disruption and looping audio aimed at users on specified language and domain combinations. This is active protestware behavior and warran...

The published entrypoint directly reaches targeted browser disruption and unsolicited political messaging in the bundled code. This is active protestware behavior, so the package should b...

The automatic postinstall installs an unsigned executable through a path explicitly designed to avoid browser-added quarantine, with additional quarantine removal on macOS. This is concre...

The source confirms that npm postinstall automatically writes bundled skills across global AI-agent skill directories. This meets the policy for blocking unconsented mutation of a broad a...

The package entrypoint immediately evaluates obfuscated code with remote loader, network, and child-process indicators, supporting a malicious verdict despite the absence of install hooks...

The package automatically changes shell profiles during global installation and redirects Anthropic clients through its tenant endpoint. This supports blocking for unconsented postinstall...

The package's install lifecycle automatically mutates user-level AI assistant skill and MCP configuration. This meets the policy for blocking unconsented lifecycle changes to an AI-agent...

The published entrypoint contains active, targeted browser disruption behavior that is unrelated to a threshold compiler’s function. This supports a malicious protestware verdict and block.

The install hook silently modifies Claude Code configuration and capabilities when Claude is detected, which is an unconsented install-time change to an AI-agent control surface. This sup...

The package entrypoint loads a heavily obfuscated module that invokes shell execution and includes network capability. This concrete, reachable behavior warrants blocking despite the exac...

The install lifecycle silently modifies global and project AI-agent control surfaces across multiple tools. This meets the blocking policy for unconsented install-time mutation of broad a...

Source inspection confirms unconsented postinstall mutation across multiple AI agent configurations, global instruction files, and shell startup files. This meets the blocking policy for...

The postinstall hook performs unconsented writes to multiple AI-agent configuration surfaces in both the project and user home. Under the install control-surface policy, this warrants blo...

The package performs unconsented postinstall writes to broad AI-agent configuration locations, establishing an active control-surface mutation attack surface. The source supports blocking...

The manifest's postinstall hook reaches code that automatically writes to multiple AI-agent configuration locations without a consent prompt. This meets the block criterion for unconsente...

The package automatically mutates global AI-agent instruction directories during postinstall, creating an unconsented agent control surface. This matches the install-hook blocking policy...

The postinstall hook reaches code that mutates broad AI-agent configuration locations without a consent prompt. This meets the blocking policy for unconsented installation-time control-su...