registry / AI-agent control-surface risk / twintest

twintest

Windows desktop application test automation framework — Appium 2.x + appium-windows-driver + WebDriverIO + Cucumber-js.

This report shows the evidence for twintest@1.0.6. 4 of 6 known versions are blocked or flagged. See all versions.

AI Security Review

scanned 23d ago · by lpm-firewall-ai

Inspected source proves unconsented postinstall mutation of a foreign, user-wide AI-agent control surface. This meets the supplied blocking policy despite runtime hook guards.

Static reason

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

Trigger

npm installation with lifecycle scripts enabled.

Impact

Persistent changes to the agent's available instructions and tool-hook behavior extend beyond this package.

Mechanism

The postinstall script copies skills and hook scripts into Claude directories and saves command-hook registrations in user settings.

Policy narrative

Installing the package activates its postinstall script, which installs skills into project and home Claude directories and registers executable hooks in home Claude settings. Runtime session checks restrict hook behavior but do not provide consent for these automatic configuration changes. The hook skip variable is an opt-out and does not stop skill installation.

AI rationale

Inspected source proves unconsented postinstall mutation of a foreign, user-wide AI-agent control surface. This meets the supplied blocking policy despite runtime hook guards.

Evidence

package.jsonscripts/postinstall.cjs.claude/skills.claude/hooks.claude/settings.json

Decision evidence

public snapshot

AI identified this as a dangerous AI-agent capability at 99.0% confidence with low false-positive risk.

Evidence for policy risk

package.json automatically runs scripts/postinstall.cjs after installation.
The script directly invokes skill and hook installation without a consent prompt.
Skills are copied into both project and user Claude skill directories.
Hook installation defaults to enabled and targets the user's Claude settings.
Command hooks are registered and the modified settings are saved.

Evidence against

Hook execution uses session activation checks.
An environment variable can skip hook installation, but skills are still installed.

Behavioral surface

Source

ChildProcessDynamicRequireEnvironmentVarsFilesystemShell

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 18 file(s), 152 KB of source, external domains: www.oracle.com

Source & flagged code

16 flagged

package.jsonView file

Lines 34-74json

34 "./wdio.conf": {

35 "types": "./dist/wdio.conf.d.ts",

36 "import": "./dist/wdio.conf.js",

37 "default": "./dist/wdio.conf.js"

38 }

39 },

40 "typesVersions": {

41 "*": {

42 "config": ["dist/config.d.ts"],

43 "hooks": ["dist/hooks.d.ts"],

44 "steps/common": ["dist/steps/common.steps.d.ts"],

45 "steps/database": ["dist/steps/database.steps.d.ts"]

46 }

47 },

48 "scripts": {

49 "test": "wdio run wdio.conf.ts",

50 "test:repair": "node bin/self-repair.mjs",

51 "appium:setup": "ts-node scripts/setup-appium.ts",

52 "appium:start": "appium --address 127.0.0.1 --port 4723 --allow-cors",

53 "appium:doctor": "appium driver doctor windows",

54 "postinstall": "node scripts/postinstall.cjs && node scripts/install-oracle-client.cjs",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L54

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg · L54

55 "build": "tsc -p tsconfig.build.json",

56 "prepublishOnly": "npm run build"

57 },

58 "bin": {

59 "twintest-self-repair": "bin/self-repair.mjs"

60 },

61 "files": [

62 "dist",

63 "bin",

64 "hooks",

65 "skills",

66 "schemas",

67 "scripts/postinstall.cjs",

68 "scripts/install-oracle-client.cjs",

69 "README.md"

70 ],

71 "dependencies": {

72 "@types/node": "^20.0.0",

73 "@wdio/appium-service": "^9.0.0",

74 "@wdio/cli": "^9.0.0",

bin/self-repair.mjsView file

Lines 1-31javascript

1#!/usr/bin/env node

2// self-repair.mjs — external driver for autonomous desktop suite repair.

3//

4// Adapted from achilles self-repair: baselines the suite, classifies failures,

5// spawns one repair worker per red feature file, verifies heals, and writes

6// an audit-grade report.

7//

8// Consumers reach this through `npm run test:repair`.

9//

10// Exit codes: 0 = every test green or explained, 2 = unresolved, 1 = driver error.

dist/config.jsView file

15*/ L16: import net from 'net'; L17: import path from 'path'; L18: import { execSync } from 'child_process'; L19: const APP_PATH = process.env.APP_PATH || 'Root'; L20: const APP_TOP_LEVEL_WINDOW = process.env.APP_TOP_LEVEL_WINDOW;

Critical

Reverse Shell

Source matches reverse-shell style process and socket wiring.

dist/config.jsView on unpkg · L15

15Trigger-reachable chain: manifest.exports -> dist/config.js L15: */ L16: import net from 'net'; L17: import path from 'path'; L18: import { execSync } from 'child_process'; L19: const APP_PATH = process.env.APP_PATH || 'Root'; L20: const APP_TOP_LEVEL_WINDOW = process.env.APP_TOP_LEVEL_WINDOW;

Critical

Trigger Reachable Dangerous Capability

A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.

dist/config.jsView on unpkg · L15

26try { L27: // Use PowerShell to test TCP connection (Windows-compatible, synchronous) L28: execSync(`powershell -NoProfile -Command "try { $c = [System.Net.Sockets.TcpClient]::new(); $c.Connect('127.0.0.1', ${port}); $c.Close(); exit 0 } catch { exit 1 }"`, { stdio: 'ign...

High

Shell

Package source references shell execution.

dist/config.jsView on unpkg · L26

dist/wdio.conf.jsView file

Lines 5-45javascript

5 *

6 * npx wdio run node_modules/twintest/dist/wdio.conf.js

7 *

8 * Or add a script to package.json:

9 *

10 * "test"

scripts/postinstall.cjsView file

Lines 1-26javascript

1#!/usr/bin/env node

3/**

4 * postinstall.js — auto-install twintest skills and hooks.

5 *

6 * Adapted from the achilles postinstall pattern:

7 * - Copies skills to ~/.claude/skills/ and <project>/.claude/skills/

8 * - Copies hooks to ~/.claude/hooks/ and registers them in settings.json

9 * - Idempotent: safe to run multiple times

10 */

hooks/winappdriver-session-guard.shView file

•path = hooks/winappdriver-session-guard.sh kind = build_helper sizeBytes = 1758 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

hooks/winappdriver-session-guard.shView on unpkg

dist/DesktopSteps.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.

dist/DesktopSteps.js#virtual:normalized:round1View on unpkg

scripts/install-oracle-client.cjsView file

•matchType = normalized_sha256 matchedPackage = twintest@1.0.5 matchedPath = scripts/install-oracle-client.cjs matchedIdentity = npm:dHdpbnRlc3Q:1.0.5 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.

scripts/install-oracle-client.cjsView on unpkg

dist/DesktopSteps.jsView file

•matchType = normalized_sha256 matchedPackage = twintest@1.0.5 matchedPath = dist/DesktopSteps.js matchedIdentity = npm:dHdpbnRlc3Q:1.0.5 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.

dist/DesktopSteps.jsView on unpkg

Source & flagged code

16 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/postinstall.cjs && node scripts/install-oracle-client.cjs

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.cjs && node scripts/install-oracle-client.cjs

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

bin/self-repair.mjsView file

11L12: import { spawn, execSync } from 'node:child_process'; L13: import {

High

Child Process

Package source references child process execution.

bin/self-repair.mjsView on unpkg · L11

Findings

3 Critical10 High5 Medium5 Low

CriticalReverse Shelldist/config.js

CriticalAi Agent Control Hijackscripts/postinstall.cjs

CriticalTrigger Reachable Dangerous Capabilitydist/config.js

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processbin/self-repair.mjs

HighShelldist/config.js

HighSame File Env Network Executiondist/config.js

HighTrigger Reachable External Ai Agent Control Surface Mutationscripts/postinstall.cjs

HighSemantic Analysis Limiteddist/DesktopSteps.js#virtual:normalized:round1

HighKnown Malware Source Similarityscripts/install-oracle-client.cjs

HighKnown Malware Source Similaritybin/self-repair.mjs

Affected versions and remediation

This report applies to twintest@1.0.6.

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

Avoid installing twintest@1.0.6. 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: 2026-09-30 09:41:59.063Z (UTC).

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

Versions

6 known versions

1.0.6latestshownMaliciousSoft block: AI-agent control surfaceBlockPublishedSep 21, 2026, 8:16 AM 1.0.5MaliciousSoft block: AI-agent control surfaceBlockPublishedSep 3, 2026, 10:36 AM 1.0.1MaliciousSoft block: AI-agent control surfaceBlockPublishedAug 27, 2026, 3:02 PM 1.0.0Scan Failed—AllowUnder ReviewAug 24, 2026, 2:18 PM

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Versions

Related packages

Other AI-agent control-surface risk packages

Behavioral surface

Related packages

Other AI-agent control-surface risk packages

Source & flagged code