registry / Remote code execution / @metaphi-ai/hum

@metaphi-ai/hum

Hum: a self-improving coding agent for your terminal.

This report shows the evidence for @metaphi-ai/hum@0.2.7. 7 of 87 known versions are blocked or flagged. The newest version, 0.2.65, is flagged. See all versions.

AI Security Review

scanned 1mo ago · by lpm-firewall-ai

This package performs unconsented install-time remote code execution through a curl-to-shell bootstrap and a second opaque package installation. That supply-chain execution path is sufficient to block publication.

Static reason

High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review

Trigger

npm installation, through the postinstall lifecycle hook

Impact

A remote party controlling the bootstrap or Python package supply chain can execute code and modify the user's home directory during npm installation.

Mechanism

Remote shell bootstrap followed by automatic Python package installation

Attack narrative

The postinstall hook invokes installCore automatically. If uv is absent, installCore runs a script fetched from astral.sh through sh or PowerShell with execution-policy bypass. It then creates ~/.hum/core and uses uv pip to download and install hum-cli, importing the result to verify it. This is an automatic remote-code bootstrap at dependency-install time, with no hash or package lock for the fetched installer or Python payload.

AI rationale

Evidence

scripts/postinstall.jsdist/install.mjs~/.hum/core~/.hum/core-install.json~/.local/bin/uv

Network endpoints2

astral.sh/uv/install.ps1astral.sh/uv/install.sh

Decision evidence

public snapshot

AI called this Malicious at 98.0% confidence as Malware with low false-positive risk.

Evidence for block

The npm postinstall hook automatically imports and calls the installer.
The installer downloads a remote script and pipes it directly to a shell or PowerShell with execution-policy bypass.
The hook creates a home-directory virtual environment and installs the separate hum-cli Python package, whose code is then imported during installation.

Evidence against

The inspected installer contains no direct credential harvesting or explicit data-exfiltration logic.

Behavioral surface

Source

ChildProcessDynamicRequireEnvironmentVarsFilesystemShell

Supply chain

HighEntropyStringsUrlStrings

Manifest

NoLicense

scanned 3 file(s), 264 KB of source, external domains: astral.sh, github.com, react.dev

Oversized source lightweight scan

dist/cli.mjs3.94 MB file, sampled 256 KB

Source & flagged code

9 flagged

package.jsonView file

Lines 1-13json

2 "name": "@metaphi-ai/hum",

3 "version": "0.2.7",

4 "description": "Hum: a self-improving coding agent for your terminal.",

5 "license": "UNLICENSED",

6 "type": "module",

7 "bin": { "hum": "bin/hum.js" },

8 "files": ["bin", "dist", "scripts"],

9 "scripts": { "postinstall": "node scripts/postinstall.js" },

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L9

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg · L9

10 "engines": { "node": ">=22" },

11 "repository": { "type": "git", "url": "git+https://github.com/metaphi-labs/humboldt.git" }

12}

dist/install.mjsView file

Lines 1-21javascript

1// src/install.js

2import { spawnSync } from "node:child_process";

3import fs from "node:fs";

High

Child Process

Package source references child process execution.

dist/install.mjsView on unpkg · L1

4import os from "node:os";

bin/hum.jsView file

Lines 1-19javascript

1#!/usr/bin/env node

2// `hum` from npm: the Ink front-end, which starts Hum's core (Python, installed by postinstall

3// via uv) and talks to it over localhost. Everything the person sees is here; everything the

4// session records is in the core.

5import { fileURLToPath, pathToFileURL } from 'node:url';

6import path from 'node:path';

dist/cli.mjsView file

485"use strict"; L486: "production" !== process.env.NODE_ENV && (function() { L487: function defineDeprecationWarning(methodName, info) { ... L624: specialPropKeyWarningShown || (specialPropKeyWarningShown = true, console.error( L625: "%s: `key` is not a prop. Trying to access it will result in `undefined` being returned. If you need to access the same value within the child component, you should pass it as a di... L626: displayName ... L1664: cwdFunction = isBrowser ? () => { L1665: throw new Error("`process.cwd()` only works in Node.js, not the browser."); L1666: } : process2.cwd; ... L1768: } L1769: const imageBuffer = Buffer.from(data); L1770: return wrapOsc(returnValue + `;size=${imageBuffer.byteLength}:` + imageBuffer.toString("base64") + BEL);

Critical

Browser Credential Phishing

Source appears to collect browser login credentials for exfiltration.

dist/cli.mjsView on unpkg · L485

88734import fs5 from "node:fs"; L88735: import net from "node:net"; L88736: import os5 from "node:os"; ... L88738: import crypto from "node:crypto"; L88739: import { spawn as spawn2, spawnSync as spawnSync3 } from "node:child_process"; L88740: function uvToolDirs() { L88741: const dirs = []; L88742: if (process.env.UV_TOOL_DIR) dirs.push(process.env.UV_TOOL_DIR); L88743: if (win3) {

High

Same File Env Network Execution

A single source file combines environment access, network access, and code or shell execution; review context before blocking.

dist/cli.mjsView on unpkg · L88734

4648import process3 from "node:process"; L4649: import { execFileSync } from "node:child_process"; L4650: import fs from "node:fs"; ... L4705: try { L4706: const statContents = fs.readFileSync("/proc/self/stat", "utf8"); L4707: const closingParenthesisIndex = statContents.lastIndexOf(") "); ... L71088: /* metaparameters */ L71089: "alias audit before loglevel noop require subscribe tag owner ensure group mode name|0 changes context force incl lens load_path onlyif provider returns root show_diff type_check e... L71090: ),

Source & flagged code

9 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/postinstall.js

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.js

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg

dist/install.mjsView file

1// src/install.js L2: import { spawnSync } from "node:child_process"; L3: import fs from "node:fs";

High

Child Process

Package source references child process execution.

dist/install.mjsView on unpkg · L1

Findings

1 Critical6 High4 Medium5 Low

CriticalBrowser Credential Phishingdist/cli.mjs

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processdist/install.mjs

HighShelldist/install.mjs

HighSame File Env Network Executiondist/cli.mjs

HighRemote Agent Bridgedist/cli.mjs

HighKnown Malware Source Similaritydist/install.mjs

MediumAmbiguous Install Lifecycle Scriptpackage.json

MediumDynamic Requirebin/hum.js

MediumEnvironment Vars

MediumStructural Risk Force Deep Review

LowScripts Present

LowFilesystem

Low

Affected versions and remediation

This report applies to @metaphi-ai/hum@0.2.7.

The versions list below shows the verdict for each recorded version.

Avoid installing @metaphi-ai/hum@0.2.7. Remove it from direct dependencies and check your lockfile for transitive copies.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.
If this version ran, investigate the affected machine and build environment. Rotate credentials it could access and rebuild from a trusted environment.

Evidence last updated: 2026-09-14 10:18:25.340Z (UTC).

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

Versions

87 known versions

0.2.65latestSuspiciousCredential ExfiltrationWarnPublishedOct 15, 2026, 1:53 AM 0.2.64SuspiciousCredential ExfiltrationWarnPublishedOct 14, 2026, 7:08 PM 0.2.63SuspiciousData ExfiltrationWarnPublishedOct 14, 2026, 6:23 PM 0.2.62SuspiciousCredential ExfiltrationWarnPublishedOct 13, 2026, 9:07 PM

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Versions

Related packages

Other remote code execution packages

Behavioral surface

Related packages

Other remote code execution packages

Source & flagged code