registry / AI-agent capability abuse / @ev-ai/agent-hook / 1.2.0

@ev-ai/agent-hook@1.2.0

Cross-runner runtime telemetry hook for Claude Code, Cursor, and Codex

AI Security Review

scanned 22d ago · by lpm-firewall-ai

Source shows explicit agent configuration mutation and subsequent automatic hook updates, warranting a capability warning under the supplied policy. No npm lifecycle attack, self-dependency chain, or confirmed credential theft was established.

Static reason

One or more suspicious static signals were detected.

Trigger

The user runs the CLI install command with a collector URL, then runs configured agent sessions.

Impact

Agent events activate telemetry code and may run updated package code, creating a guarded agent integration risk.

Mechanism

The CLI merges hook commands into project configuration, and the launcher selects cached hook artifacts and schedules refreshes unless disabled.

AI rationale

Evidence

dist/cli.mjsdist/hook.mjsdist/launcher.mjsdist/launcher-update.mjs.claude/settings.json.cursor/hooks.json.codex/hooks.json

Network endpoints2

hooks.ev-ai.airegistry.npmjs.org

Decision evidence

public snapshot

AI called this Suspicious at 91.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The explicit CLI install command activates agent hook configuration changes.
dist/cli.mjs merges telemetry commands into agent hook settings, including .cursor/hooks.json.
Configured agent sessions can automatically launch a detached refresh process; environment or saved preferences can disable updates.
The updater restricts package archive URLs to this package on registry.npmjs.org and checks downloaded integrity.
dist/hook.mjs posts telemetry envelopes to the selected collector; the indicated configuration helper only classifies paths.
package.json exposes the launcher command and has no npm lifecycle scripts or runtime self-dependencies.

Evidence against

Configuration changes require an explicit CLI command rather than automatic npm installation.
No concrete malicious execution or covert credential forwarding was established.

Behavioral surface

Source

ChildProcessDynamicRequireEnvironmentVarsFilesystemNetwork

Source & flagged code

7 flagged

dist/launcher.mjsView file

Lines 1-6javascript

1#!/usr/bin/env node

3import{spawn as Hn}from"node:child_process";import{homedir as Bn}from"node:os";import{fileURLToPath as Rt}from"node:url";import{createHash as Jt}from"node:crypto";import{join as J}from"node:path";var Ct=/^(\d+)\.(\d+)\.(\d+)(?:-([0-9A-Za-z.-]+))?$/,kt=/^>=(\d+)(?:\.(\d+)(?:\.(\d+))?)?$/,Nt=/^v?(\d+)\.(\d+)\.(\d+)/;function be(e){let t=Ct.exec(e);if(t)return{...q(t),prerelease:t[4]?.split(".")}}function Ot(e,t){let n=be(e),r=be(t);if(!(!n||!r))return n.major!==r.major?n.major-r.major:n.minor!==r.minor?n.minor-r.minor:n.patch!==r.patch?n.patch-r.patch:Tt(n.prerelease,r.prerelease)}function Ee(e, ...

4`)}catch{}return 1}if(n==="bundled_cli")return(await wt(t)).main(e);if(n==="refresh"||n==="status")try{let s=await t.readPackageInfo().catch(()=>At),i=await ve(t);return i?n==="refresh"?await i.runRefresh(e.slice(1),t.env,t.cwd,t.home,{packageVersion:s.version}):await i.runStatus(e,t,s):0}catch{return 0}let r=()=>{},o=new Promise(s=>{r=s}),c={...t,exit:s=>{t.exit(s),r(s)}};return Promise.race([Kn(e,c,r),o])}async function Kn(e,t,n){let r=ct(t),o=Et(t.stdin,t),c,s=()=>c??=lt(e,o.classifiableText(),t.env);o.whenSettled().then(()=>r.onEventClassified(s().isInProcess));let i=!1,a=new Promise(p=>{r ...

High

Child Process

Package source references child process execution.

dist/launcher.mjsView on unpkg · L2

High

Cross File Remote Execution Context

Source spawns a local helper that also contains network and dynamic execution context; review data flow before blocking.

dist/launcher.mjsView on unpkg · L2

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/launcher.mjsView on unpkg · L2

5`)}catch{}process.exit(1)}try{process.stdout.write('{"continue":true}')}catch{}process.exit(0)});

Long lines were clipped for display.

dist/hook.mjsView file

Lines 1-32javascript

1import { createRequire as __createRequire } from 'node:module';

2const require = __createRequire(import.meta.url);

4var gc=Object.create;var

Source & flagged code

7 flagged · loading source

dist/launcher.mjsView file

2L3: import{spawn as Hn}from"node:child_process";import{homedir as Bn}from"node:os";import{fileURLToPath as Rt}from"node:url";import{createHash as Jt}from"node:crypto";import{join as J}... L4: `)}catch{}return 1}if(n==="bundled_cli")return(await wt(t)).main(e);if(n==="refresh"||n==="status")try{let s=await t.readPackageInfo().catch(()=>At),i=await ve(t);return i?n==="ref...

High

Child Process

Package source references child process execution.

dist/launcher.mjsView on unpkg · L2

2Cross-file remote execution chain: dist/launcher.mjs spawns dist/cli.mjs; helper contains network access plus dynamic code execution. L2: L3: import{spawn as Hn}from"node:child_process";import{homedir as Bn}from"node:os";import{fileURLToPath as Rt}from"node:url";import{createHash as Jt}from"node:crypto";import{join as J}... L4: `)}catch{}return 1}if(n==="bundled_cli")return(await wt(t)).main(e);if(n==="refresh"||n==="status")try{let s=await t.readPackageInfo().catch(()=>At),i=await ve(t);return i?n==="ref...

High

Cross File Remote Execution Context

Source spawns a local helper that also contains network and dynamic execution context; review data flow before blocking.

dist/launcher.mjsView on unpkg · L2

Findings

6 High4 Medium4 Low

HighChild Processdist/launcher.mjs

HighSame File Env Network Executiondist/hook.mjs

HighCommand Output Exfiltrationdist/hook.mjs

HighSandbox Evasion Gated Capabilitydist/hook.mjs

HighCloud Metadata Accessdist/hook.mjs

HighCross File Remote Execution Contextdist/launcher.mjs

MediumDynamic Requiredist/launcher.mjs

MediumNetwork

MediumEnvironment Vars

MediumStructural Risk Force Deep Review

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

LowNo License

Affected versions and remediation

This report applies to @ev-ai/agent-hook@1.2.0.

See all versions of @ev-ai/agent-hook for other recorded verdicts.

Review the evidence and your use of @ev-ai/agent-hook@1.2.0 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: 2026-10-01 08:21:24.121Z (UTC).

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