registry / @mingliuyiming/qianfan-cli-test / 1.0.0

@mingliuyiming/qianfan-cli-test@1.0.0

Qianfan command-line client

AI Security Review

scanned 1mo ago · by lpm-firewall-ai

LPM flags this version as an AI-agent control-surface risk. Installation fetches and executes a remote native binary. A global install then non-interactively invokes that binary to install skills into detected local Agent directories.

Static reason

One or more suspicious static signals were detected.

Trigger

npm global installation, unless the package-specific skip variables are set.

Impact

An untrusted package release can place and execute native code, then modify a broad local AI-agent control surface without an explicit setup command.

Mechanism

Postinstall remote binary staging followed by automatic Agent skill installation.

Policy narrative

The postinstall hook selects a manifest URL, downloads a native executable, verifies the package-provided hash, saves it under bin, and makes it executable. During global installation it automatically runs the opaque executable with +connect --yes; source comments state this installs skills into detected Agent directories. The package name differs from the identity used throughout its code and documentation, increasing the risk that this lifecycle behavior is impersonating another package.

AI rationale

This is unconsented postinstall execution of an opaque remote binary that automatically modifies a broad AI-agent control surface on global installs. The checksum does not establish trust because it is supplied by the same package release.

Evidence

package.jsonscripts/postinstall.jsmanifest.jsonbin

Network endpoints2

aipe-dragonboat.cdn.bcebos.com

@mingliuyiming/qianfan-cli-test@1.0.0

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Source & flagged code