registry / AI-agent control-surface risk / @chem-x/starter-kit / 26.10.4-1258

@chem-x/starter-kit@26.10.4-1258

Chemical X Protocol: Private drop-in architecture starter kit and capsule generator

AI Security Review

scanned 11d ago · by lpm-firewall-ai

The manifest activates a postinstall that writes a package-owned MCP server into consumer agent configuration without consent. This meets the policy for unconsented install-time mutation of an AI-agent control surface.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

Installing the package through npm in a consumer project.

Impact

The install changes agent configuration and enables the package server to appear as an agent capability; inspected source does not establish credential theft or exfiltration.

Mechanism

The postinstall calls the MCP installer, which creates or updates Cursor and VS Code MCP configuration files without an explicit consent gate.

Policy narrative

During npm installation, the postinstall selects the consumer project directory and invokes the MCP configuration installer. That installer creates or updates project configuration for Cursor and VS Code without asking for consent, registering the Chemical X MCP server. The package also includes agent guidance directing agents toward its own tools and prohibiting ordinary file inspection methods. The inspected source supports an unconsented install-time agent control-surface mutation; it does not establish credential theft or exfiltration.

AI rationale

Evidence

package.jsoncli/postinstall.jscli/mcp/installer.jscli/installer.js.cursor/mcp.json.vscode/mcp.jsonAGENTS.md

Decision evidence

public snapshot

AI identified this as a dangerous AI-agent capability at 98.0% confidence with low false-positive risk.

Evidence for policy risk

The package postinstall hook runs a consumer-project MCP configuration installer without a consent check.
The installer writes a Chemical X server entry into Cursor and VS Code MCP configuration files during installation.
The package adds agent directives that steer agents toward its own tools and limit native file inspection.
The installer tests confirm that project MCP configuration and package scripts are created.

Evidence against

The postinstall skips when run inside the starter-kit repository itself.
The MCP setup writes configuration and helper scripts; inspected source does not show credential collection or network exfiltration.

Behavioral surface

Source

ChildProcessEnvironmentVarsFilesystemNetworkShell

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 327 file(s), 1.73 MB of source, external domains: 127.0.0.1, api.github.com, api.local, chemicalx.xophz.com, example.com, fallback.com, github.com, img.shields.io, mycompassconsulting.com, randomsite.com, reports.chemicalx.xophz.com, www.w3.org

Source & flagged code

11 flagged

package.jsonView file

Lines 35-75json

35 ".": "./cli/index.js",

36 "./audit": "./cli/audit.js",

37 "./build": "./cli/build.js",

38 "./search": "./cli/search.js",

39 "./mcp": "./cli/mcp/index.js",

40 "./reader": "./cli/reader.js",

41 "./patcher": "./cli/patcher.js",

42 "./errors": "./cli/errors/index.js"

43 },

44 "scripts": {

45 "chemx": "node cli/index.js",

46 "bench": "node benchmarks/run-benchmark.mjs",

47 "ui": "node cli/index.js ui",

48 "dev:ui": "node cli/index.js ui --dev",

49 "dev:ui:watch": "node --watch --watch-path=cli cli/index.js ui --dev",

50 "test": "node --test cli/audit/*.spec.js cli/mcp/*.spec.js cli/build/*.spec.js cli/team/*.spec.js cli/errors/*.spec.js cli/*.spec.js",

51 "check:frameworks": "node scripts/check-framework-generation.mjs",

52 "prepublishOnly": "node scripts/check-framework-generation.mjs",

53 "typecheck": "tsc --noEmit",

54 "publish:both": "node scripts/publish-both.mjs",

55 "postinstall": "node cli/postinstall.js",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L55

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg · L55

56 "chemx:mcp": "chemx mcp",

57 "chemx:verify": "chemx verify",

58 "chemx:test": "chemx test",

59 "chemx:typecheck": "chemx typecheck"

60 },

61 "dependencies": {

62 "@babel/parser": "^7.29.8",

63 "@babel/traverse": "^7.29.8",

64 "@babel/types": "^7.29.8"

65 },

66 "devDependencies": {

67 "@types/babel__traverse": "^7.28.0",

68 "@types/node": "^22.20.2",

69 "@types/react": "^18.3.31",

70 "@types/react-dom": "^18.3.7",

71 "chemx": "26.9.20-408",

72 "react": "^18.3.1",

73 "react-dom": "^18.3.1",

74 "typescript": "^5.9.3"

75 }

cli/errors/errors.spec.jsView file

Lines 7-47javascript

7 maskSensitiveTokens,

8 normalizeHomePath,

9 sanitizeText,

10 sanitizeStackTrace

11} from './sanitizer.js';

12import {

13 buildIssueTitle,

14 buildIssueBody,

15 buildIssueWebUrl,

16 formatIssueContent

cli/audit-preflight-git.jsView file

Lines 1-21javascript

1import fs from 'node:fs';

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

High

Child Process

Package source references child process execution.

cli/audit-preflight-git.jsView on unpkg · L1

4const SOURCE_FILE_PATTERN = /

cli/build/executor.jsView file

Lines 1-31javascript

1import { spawn } from 'node:child_process';

3export const executeBuild = (command, cwd = process.cwd(), options = {})

cli/license.jsView file

Lines 1-23javascript

1import fs from 'node:fs';

2import path from 'node:path';

3import os from 'node:os';

4import { spawnSync } from 'node:child_process';

cli/mcp/installer.jsView file

1Install-time code directly mutates a foreign AI-agent control surface: L20: export const mergeMcpServerConfig = (existingJsonString = '', serverDef = {}) => { L21: let parsed = { mcpServers: {} }; L22: ... L30: L31: const hasMcpServersObject = Boolean(parsed.mcpServers && typeof parsed.mcpServers === 'object'); L32: if (!hasMcpServersObject) { L33: parsed.mcpServers = {}; L34: } L35: L36: parsed.mcpServers[SERVER_KEY] = serverDef; L37: return JSON.stringify(parsed, null, 2) + '\n'; ... L90: const cursorDir = path.join(resolvedTarget, '.cursor'); Write operation from cli/mcp/installer.js: L20: export const mergeMcpServerConfig = (existingJsonString = '', serverDef = {}) => { L21: let parsed = { mcpServers: {} }; L22: ... L30: L31: const hasMcpServersObject = Boolean(parsed.mcpServers &

Critical

Ai Agent Control Hijack

Source creates an unconsented AI-agent control surface through install-time mutation or a default unauthenticated remote skill channel.

cli/mcp/installer.jsView on unpkg · L1

cli/pillars-wizard.jsView file

•Manifest-trigger-reachable source links an external AI-agent control path to a behavior-bearing write operation. st pick = usegum ? gumchoose(choices, 'select architectural pillars configuration') : await promptquestion('select pillars: [1=recommended, 2=strict, 3=custom, 4=minimal, 5=none] (default: 1): '); if (pick && (pick.startswith('2.') || pick === '2')) { selectedpresetkey = 'strict'; selectedpillarids = [...pillar_presets.strict.pillars]; } else if (pick && (pick.startswith('3.') || pick === '3')) { selectedpresetkey = 'custom'; for (const pillar of pillars) { const defaultchoice = pillar_presets.recommended.pillars.includes(pillar.id) ? 'y/n' : 'y/n'; const isdefaultyes = defaultchoice === 'y/n'; const answer = a

High

Trigger Reachable External Ai Agent Control Surface Mutation

Manifest-trigger-reachable source writes behavior-bearing configuration into a user or project AI-agent control surface.

cli/pillars-wizard.jsView on unpkg

scripts/pre-commit.shView file

•path = scripts/pre-commit.sh kind = build_helper sizeBytes = 8215 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

scripts/pre-commit.shView on unpkg

cli/audit/ai-slop-detector.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 2

High

Semantic Analysis Limited

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

cli/audit/ai-slop-detector.js#virtual:normalized:round1View on unpkg

Source & flagged code

11 flagged · loading source

package.jsonView file

•scripts.postinstall = node cli/postinstall.js

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node cli/postinstall.js

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

cli/errors/errors.spec.jsView file

27patternName = github_pat severity = critical line = 27

Critical

Critical Secret

Package contains a critical-looking secret pattern.

cli/errors/errors.spec.jsView on unpkg · L27

Findings

3 Critical6 High5 Medium5 Low

CriticalCritical Secretcli/errors/errors.spec.js

CriticalAi Agent Control Hijackcli/mcp/installer.js

CriticalSecret Patterncli/errors/errors.spec.js

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processcli/audit-preflight-git.js

HighShellcli/build/executor.js

HighSame File Env Network Executioncli/license.js

HighTrigger Reachable External Ai Agent Control Surface Mutationcli/pillars-wizard.js

HighSemantic Analysis Limitedcli/audit/ai-slop-detector.js#virtual:normalized:round1

MediumAmbiguous Install Lifecycle Scriptpackage.json

MediumNetwork

Medium

Affected versions and remediation

This report applies to @chem-x/starter-kit@26.10.4-1258.

See all versions of @chem-x/starter-kit for other recorded verdicts.

Avoid installing @chem-x/starter-kit@26.10.4-1258. 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: 2026-10-12 06:01:21.518Z (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

Related packages

More from @chem-x

Other AI-agent control-surface risk packages

Behavioral surface

Related packages

More from @chem-x

Other AI-agent control-surface risk packages

Source & flagged code