Evidence for AI clean decision
- The package describes its purpose as human approval and answers for AI-agent sessions.
- The manifest has no install, preinstall, or postinstall hook.
- Setup requires a user-provided or deliberately paired Pushary API key; the Claude wrapper uses that key to poll its agent service and forwards returned text into the active session.
Evidence against
- The wrapper can receive instructions from the Pushary service while a session is active, which is a powerful capability but matches the package’s stated remote-approval purpose.
- The setup bundle includes a shell alias that routes Claude through Pushary; the inspected code describes it as reversible with `pushary clean`.
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 155 files · 2.45 MB of source
9 external domains referenced
- claude.ai
- discord.gg
- fx.sh
- github.com
- nodejs.org
- pushary.com
- registry.npmjs.org
- schemas.microsoft.com
- www.apple.com