OpenAgents Launcher — install, configure, and run AI coding agents from your terminal
Allowed — no malicious behavior detected
17 low-signal patterns reviewed and cleared.
AI security review
scanned · by lpm-firewall-ai
The manifest has no npm lifecycle hook, and the inspected runtime excerpts are consistent with the package’s launcher function. Scanner flags alone do not establish malicious behavior.
Using the CLI or invoking an agent adapter, workspace tool, or explicit autostart command.
Impact
No unrelated credential transfer, unconsented persistence, or other malicious effect was established.
Mechanism
The reviewed code reads local npm configuration, sends configured API requests, communicates with agent processes, and writes workspace skill files.
Decision evidence
public snapshot
Evidence for AI clean decision
package.json declares CLI entrypoints and development scripts, with no npm lifecycle hook.
The Claude and Cursor adapter excerpts write workspace skill files using configured workspace values; the cited code does not show an unrelated configuration target.
The inspected credential-related paths read npm configuration for binary discovery and registry selection; Codex direct requests send a configured API key to the selected direct API URL.
The WSL and adapter runtime excerpts concern cache refresh, WebSocket request headers, task continuation, subprocess output, and agent responses.
The package’s adapter, workspace tool, terminal UI, installer, and autostart source aligns with its launcher purpose; no concrete attack behavior was established in the reviewed excerpts.
The dependency evidence identifies blessed as a terminal UI library and ws as a WebSocket library; the inspected source excerpts show their library entrypoints and terminal or WebSocket behavior.
Evidence against
Static scanner labels for remote execution, persistence, and child processes are not proof of malicious behavior; inspected excerpts do not establish a malicious trigger or effect.
Behavioral surface
SourceChild process · Crypto · Dynamic require · Environment variables · Filesystem · Network · Shell · Web socket
Supply chainHigh entropy strings · URL strings
ManifestNo manifest risk signals triggered.
Scanned 88 files · 1.90 MB of source
28 external domains referenced
aider.chat
ampcode.com
antigravity.google
api.anthropic.com
api.deepseek.com
api.github.com
api.moonshot.ai
api.openai.com
astral.sh
cdn.npmmirror.com
claude.ai
codearts.huaweicloud.com
cursor.com
example.com
generativelanguage.googleapis.com
github.com
mirrors.aliyun.com
mirrors.cloud.tencent.com
mirrors.huaweicloud.com
nodejs.org
opencode.ai
openrouter.ai
raw.githubusercontent.com
registry.npmjs.org
registry.npmmirror.com
workspace-endpoint.openagents.org
workspace.openagents.org
www.apple.com
Findings
3 critical · 5 high · 5 medium · 4 low
Automated scan signals; the review checked them and cleared this version. Where they are, most severe file first. Code opens a file's flagged lines, with their evidence, below.