alvin-bot@5.105.0
Self-hosted autonomous AI agent on Telegram, Slack, Discord, WhatsApp, Signal, terminal & web. Claude Agent SDK + multi-provider engine with auto-failover, detached sub-agents, zero-config memory. Local-first, telemetry-free.
AI Security Review
scanned 2mo ago · by lpm-firewall-aiNo automatic install hook or confirmed exfiltration was found, but deliberate runtime obfuscation combined with agent capabilities prevents a clean classification. The observed persistence is explicitly user-invoked rather than covert.
Decision evidence
public snapshot- The distributed runtime is deliberately obfuscated, preventing meaningful audit of all shipped execution paths.
- The bundled email plugin exposes tools that read inbox messages and send email through the local himalaya client.
- Explicit CLI commands can create persistent macOS LaunchAgents for the bot, including a watchdog.
- The browser CLI evaluates user-supplied JavaScript in a Chromium page.
- package.json has only a prepublishOnly hook; npm installation does not automatically run package code.
- Persistence and external-tool installation are routed through named CLI commands rather than an npm install hook.
- No concrete credential exfiltration endpoint or hidden network payload was found in the inspected source.
Source & flagged code
20 flagged · loading sourcePackage source references child process execution.
claude-bridge/src/claude.jsView on unpkg · L2Source file is highly similar to a previously finalized malicious package; route for source-aware review.
claude-bridge/src/claude.jsView on unpkgPackage source references dynamic require/import behavior.
claude-bridge/src/claude.jsView on unpkg · L2Findings
Affected versions and remediation
This report applies to alvin-bot@5.105.0.
See all versions of alvin-bot for other recorded verdicts.
- Review the evidence and your use of alvin-bot@5.105.0 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.