registry / AI-agent capability abuse / alvin-bot / 5.96.0

alvin-bot@5.96.0

Open-source, 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 1mo ago · by lpm-firewall-ai

No concrete malware behavior or automatic installation mutation was found. Warning is warranted because the package combines opaque runtime code with high-impact autonomous-agent capabilities.

Static reason

One or more suspicious static signals were detected.; source matched previously finalized malicious package; routed for review

Trigger

A user runs the CLI start, launchd, agent, or email features.

Impact

A user who grants the bot access could enable broad local-agent actions and persistence; obfuscation prevents fully confirming the runtime limits.

Mechanism

The runtime can spawn configured executables, execute tool commands, send email through himalaya, and install a LaunchAgent after an explicit CLI command.

AI rationale

No concrete malware behavior or automatic installation mutation was found. Warning is warranted because the package combines opaque runtime code with high-impact autonomous-agent capabilities.

Evidence

bin/cli.jsclaude-bridge/src/claude.jsdist/providers/tool-executor.jsplugins/email/index.js

Decision evidence

public snapshot

AI called this Suspicious at 79.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

Published runtime files are intentionally obfuscated, limiting static verification of their effective behavior.
The tool executor imports process-execution APIs and invokes a command safety check before execution.
The explicit start command can install a macOS LaunchAgent and a watchdog that restarts the bot.
The Claude bridge spawns a configurable external Claude executable with inherited environment variables.
The email plugin executes the installed himalaya command to send user-supplied email content.

Evidence against

package.json has no npm preinstall, install, or postinstall hook.
Persistence is reached through explicit CLI start or launchd commands, not package installation.
The inspected executor artifact includes command and protected-write guard calls.
No hidden credential export, unrelated endpoint, destructive install action, or self-dependency was found.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsEval

Source & flagged code

19 flagged

claude-bridge/src/claude.jsView file

Lines 1-22javascript

1'use strict';

3const { spawn } = require('child_process');

4const crypto = require('crypto');

High

Child Process

Package source references child process execution.

claude-bridge/src/claude.jsView on unpkg · L2

Medium

Dynamic Require

Package source references dynamic require/import behavior.

claude-bridge/src/claude.jsView on unpkg · L2

6const CLAUDE_BIN = process.env.CLAUDE_BIN || 'claude';

Low

Weak Crypto

Package source references weak cryptographic algorithms.

claude-bridge/src/claude.jsView on unpkg · L2

8// --- Stable alias per CLI session ---

9// Each session keeps one alias so resumed requests don't accumulate different

10// names in the Claude context (which would leak un-replaced aliases in output).

11const PREFIXES = ['Chat', 'Dev', 'Run', 'Ask', 'Net', 'App', 'Zen', 'Arc', 'Dot', 'Amp', 'Hex', 'Orb', 'Elm', 'Oak', 'Sky'];

12const SUFFIXES = ['Kit', 'Box', 'Pod', 'Hub', 'Lab', 'Ops', 'Bay', 'Tap', 'Rim', 'Fog', 'Dew', 'Fin', 'Gem', 'Jet', 'Cog'];

13const pick = (arr) => arr[Math.floor(Math.random() * arr.length)];

14const sessionAliasMap = new Map(); // sessionId → { alias, aliasLower, lastUsed }

15const sessionTokenMaps = new Map(); // sessionId → Map<original, replacement>

17function getSessionAlias(sessionId) {

18 if (!sessionId) {

19 const alias = pick(PREFIXES) + pick(SUFFIXES);

20 return { alias, aliasLower: alias.toLowerCase() };

21 }

22 let entry = sessionAliasMap.get(sessionId);

dist/providers/tool-executor.jsView file

Lines 1-1javascript

1const _0xd55ec7=_0x21be,_0xa47b71=_0x21be;(function(_0x15f91b,_0x5eee54){const _0x3e6b62=_0x21be,_0x5d35eb=_0x21be,_0xdfc9aa=_0x15f91b();while(

bin/cli.jsView file

Lines 2-42javascript

3/**

4 * Alvin Bot CLI — Setup, manage, and chat with your AI agent.

5 *

6 * Usage:

7 * alvin-bot setup — Interactive setup wizard

8 * alvin-bot tui — Terminal chat UI

9 * alvin-bot doctor — Check configuration

10 * alvin-bot audit — Security health check

11 * alvin-bot update — Pull latest & rebuild

12 * alvin-bot start — Start the bot

dist/web/server.jsView file

1const _0x540a63=_0x32ae,_0x3c9aee=_0x32ae;(function(_0x4eb446,_0x5cd31b){const _0x2ad2fa=_0x32ae,_0x38052e=_0x32ae,_0x23f9a9=_0x4eb446();while(!![]){try{const _0x19fc45=-parseInt(_...

High

Command Output Exfiltration

Source combines command execution, command-output handling, and outbound requests; review data flow before blocking.

dist/web/server.jsView on unpkg · L1

1Trigger-reachable command-output exfiltration chain: scripts.start -> dist/index.js -> dist/web/server.js L1: const _0x540a63=_0x32ae,_0x3c9aee=_0x32ae;(function(_0x4eb446,_0x5cd31b){const _0x2ad2fa=_0x32ae,_0x38052e=_0x32ae,_0x23f9a9=_0x4eb446();while(!![]){try{const _0x19fc45=-parseInt(_...

High

Trigger Reachable Command Output Exfiltration

A manifest entrypoint or package-local install chain reaches command-output exfiltration behavior.

dist/web/server.jsView on unpkg · L1

dist/middleware/auth.jsView file

1function _0x154d(){const _0x3a228e=['[redacted]','DhjPBq','y2fSBgjHy2TrDwvYEq','y29Kzq','mND4rwrizW','[redacted]...

High

Obfuscated Payload Loader

Source contains an obfuscated payload loader that reconstructs and executes hidden code.

dist/middleware/auth.jsView on unpkg · L1

install.shView file

•path = install.sh kind = build_helper sizeBytes = 10656 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

install.shView on unpkg

claude-bridge/src/server.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 15

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

claude-bridge/src/server.js#virtual:normalized:round1View on unpkg

plugins/email/index.jsView file

•matchType = normalized_sha256 matchedPackage = alvin-bot@5.149.0 matchedPath = plugins/email/index.js matchedIdentity = npm:YWx2aW4tYm90:5.149.0 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High

Known Malware Source Similarity

Source file is highly similar to a previously finalized malicious package; route for source-aware review.

plugins/email/index.jsView on unpkg

Source & flagged code

19 flagged · loading source

claude-bridge/src/claude.jsView file

2L3: const { spawn } = require('child_process'); L4: const crypto = require('crypto');

High

Child Process

Package source references child process execution.

claude-bridge/src/claude.jsView on unpkg · L2

2L3: const { spawn } = require('child_process'); L4: const crypto = require('crypto');

Medium

Dynamic Require

Package source references dynamic require/import behavior.

claude-bridge/src/claude.jsView on unpkg · L2

2L3: const { spawn } = require('child_process'); L4: const crypto = require('crypto'); L5: L6: const CLAUDE_BIN = process.env.CLAUDE_BIN || 'claude'; L7: ... L106: const SCRUB_WHITELIST = new Set([ L107: 'API_KEY', 'API_KEYS', 'API_URL', 'BASE_URL', 'BASE64', L108: 'HTTP_GET', 'HTTP_POST', 'HTTP_PUT', 'HTTP_DELETE', 'HTTP_PATCH', ... L294: L295: // Idle timeout: reset on every stdout activity. L296: // Claude is alive as long as it produces output (tool calls, results, etc.)

Low

Weak Crypto

Package source references weak cryptographic algorithms.

claude-bridge/src/claude.jsView on unpkg · L2

Findings

15 High6 Medium7 Low

HighChild Processclaude-bridge/src/claude.js

HighShelldist/providers/tool-executor.js

HighEntrypoint Build Divergencebin/cli.js

HighSame File Env Network Executionbin/cli.js

HighCommand Output Exfiltrationdist/web/server.js

HighSandbox Evasion Gated Capabilitybin/cli.js

HighObfuscated Payload Loaderdist/middleware/auth.js

HighCross File Remote Execution Contextbin/cli.js

HighTrigger Reachable Command Output Exfiltrationdist/web/server.js

HighTrigger Reachable Persistencebin/cli.js

HighRuntime Package Installbin/cli.js

HighObfuscated

Affected versions and remediation

This report applies to alvin-bot@5.96.0.

See all versions of alvin-bot for other recorded verdicts.

Review the evidence and your use of alvin-bot@5.96.0 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: 2026-09-22 05:01:03.324Z (UTC).

Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code