registry / grix-connector / 3.26.0

grix-connector@3.26.0

Connect local AI coding agents (Claude, Codex, Gemini, Qwen, DeepSeek, Cursor, OpenCode, Pi, OpenHuman, Reasonix) to the Grix scheduling platform. Also serves as an OpenClaw plugin for Grix channel transport.

AI Security Review

scanned 2mo ago · by lpm-firewall-ai

The package has real, risky persistence and runtime AI-agent safety-setting mutations, but source inspection does not establish malicious intent or an unconsented broad foreign control-surface write at install time.

Static reason

One or more suspicious static signals were detected.; previous stored version diff introduced dangerous source

Trigger

npm install; then user starts a Grix Claude connector session.

Impact

Reduces Claude safety confirmation and creates executable local upgrade-management state.

Mechanism

package-owned lifecycle shim plus runtime Claude configuration mutation

AI rationale

Evidence

package.jsonscripts/install-guardian.mjsscripts/upgrade-guardian.shdist/adapter/claude/claude-adapter.js~/.grix/bin/upgrade-guardian.sh~/.claude/settings.json~/.claude.json

Network endpoints1

127.0.0.1:${HEALTH_PORT}/healthz

Decision evidence

public snapshot

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

Evidence for warning

package.json invokes postinstall.
scripts/install-guardian.mjs force-copies an executable into ~/.grix/bin.
dist/adapter/claude/claude-adapter.js sets Claude's skip-dangerous-permission flag.
Claude adapter pre-trusts workspaces and injects MCP/status-line settings.

Evidence against

Postinstall writes only the package-owned ~/.grix area.
Guardian only performs local health checks and package rollback.
No credential harvesting or unrelated outbound exfiltration was confirmed.
Claude mutations occur when the connector starts a Claude session, not at npm install.

Behavioral surface

Source

ChildProcessCryptoEnvironmentVarsFilesystemNativeBindingsNetworkShellWebSocket

Supply chain

HighEntropyStringsMinifiedUrlStrings

ManifestNo manifest risk signals triggered.

Source & flagged code

11 flagged

package.jsonView file

Lines 3-43json

3 "version": "3.26.0",

4 "description": "Connect local AI coding agents (Claude, Codex, Gemini, Qwen, DeepSeek, Cursor, OpenCode, Pi, OpenHuman, Reasonix) to the Grix scheduling platform. Also serves as an OpenClaw plugin for Grix channel transport.",

5 "type": "module",

6 "main": "dist/index.js",

7 "bin": {

8 "grix-connector": "dist/grix.js"

9 },

10 "files": [

11 "dist",

12 "!dist/**/*.map",

13 "!dist/**/*.d.ts",

14 "!dist/**/*.ts",

15 "openclaw-plugin",

16 "scripts/install-guardian.mjs",

17 "scripts/analyze-audit-replays.mjs",

18 "scripts/install-guardian.sh",

19 "scripts/upgrade-guardian.sh",

20 "openclaw.plugin.json"

21 ],

22 "scripts": {

23 "postinstall": "node scripts/install-guardian.mjs",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L23

24 "prepublishOnly": "npm run build:all",

25 "build": "npm run check:isolation && node -e \"const fs=require('fs'); for(const dir of ['dist/default-skills','dist/kimi-plugins']) fs.rmSync(dir,{recursive:true,force:true})\" && tsc && node -e \"const fs=require('fs'); for(const dir of ['default-skills','kimi-plugins']) fs.cpSync('src/'+dir,'dist/'+dir,{recursive:true,filter:s=>!s.endsWith('.ts')})\" && node -e \"require('fs').cpSync('src/core/proxy/hermes-profile-relay.py','dist/core/proxy/hermes-profile-relay.py')\"",

26 "check:isolation": "node scripts/check-adapter-isolation.mjs",

27 "build:plugin": "node scripts/build-plugin.mjs",

28 "build:all": "npm run build && node scripts/minify-dist.mjs && npm run build:plugin",

29 "dev": "npm run build && node dist/grix.js",

30 "test": "node --experimental-vm-modules node_modules/vitest/vitest.mjs run",

31 "test:watch": "node --experimental-vm-modules node_modules/vitest/vitest.mjs",

32 "test:e2e": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e",

33 "test:e2e:smoke": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/smoke.test.ts",

34 "test:e2e:direct-relay": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/direct-relay-live.test.ts",

35 "test:e2e:codex-developer-instructions": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/codex-developer-instructions-protocol.test.ts",

36 "test:e2e:claude-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/claude-audit-replay.test.ts",

37 "test:e2e:codex-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/codex-audit-replay.test.ts",

38 "test:e2e:native-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/native-event-audit-replay.test.ts",

39 "test:e2e:kimi-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/native-event-audit-replay.test.ts -t \"kimi:\"",

40 "test:e2e:pi-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/pi-audit-replay.test.ts",

41 "test:e2e:reasonix-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/reasonix-audit-replay.test.ts",

42 "test:e2e:codewhale-audit": "RUN_REAL_AGENT_E2E=1 node --experimental-vm-modules node_modules/vitest/vitest.mjs run tests/e2e/codewhale-audit-replay.test.ts",

43 "audit:analyze": "node scripts/analyze-audit-replays.mjs",

openclaw-plugin/index.jsView file

Lines 3038-3078javascript

3038function formatCommandFailure(result) {

3039 const parts = [String(result.stderr ?? "").trim(), String(result.stdout ?? ""

dist/core/proxy/crypto/certificate.jsView file

Lines 1-6javascript

1import{createHash as x,createPrivateKey as G,createPublicKey as V,generateKeyPairSync as I,sign as z}from"node:crypto";import{existsSync as b}from"node:fs";import{mkdir as J,readFile as

dist/adapter/openhuman/openhuman-adapter.jsView file

Lines 1-9javascript

1import{EventEmitter as f}from"node:events";import{stat as g}from"node:fs/promises";import{mkdirSync as _}from"node:fs";import{homedir as u}from"node:os";

dist/core/provider-quota/kiro.jsView file

Lines 1-2javascript

1import{readFile as _}from"node:fs/promises";import{existsSync as m}from"node:fs";import{homedir as h}from"node:os";import{join as u}from"node:path";

dist/core/installer/npm-registry.jsView file

Lines 1-3javascript

1import{execFile as p}from"node:child_process";import{promisify as y}from"node:util";import{log as g}from"../log/logger.js";import{probeUrls as w}from"./speed-test.js";

dist/adapter/claude/claude-adapter.jsView file

Lines 1-21javascript

1import{spawn as Z,spawnSync as ee,execSync as te,execFile as We}from"node:child_process";import{promisify as Qe}from"node:util";import{randomUUID as J}from"node:crypto";import

dist/adapter/codex/codex-bridge.jsView file

1Trigger-reachable command-output exfiltration chain: manifest.main -> dist/index.js -> dist/adapter/codex/index.js -> dist/adapter/codex/codex-bridge.js L1: import{execFileSync as G}from"node:child_process";import{createInterface as le}from"node:readline";import{EventEmitter as X}from"node:events";import{stat as J}from"node:fs/promises... L2: \u2026(truncated)`,ie=80,Xe=120*1e3,Je=[{id:"gpt-5.3-codex",displayName:"GPT-5.3 Codex",defaultReasoningEffort:null,supportedReasoningEfforts:[],defaultServiceTier:null,serviceTier... L3: `)})}catch{return!1}}getStatus(){return{alive:this.alive,busy:this.currentTurnId!==null||this.compacting,sessions:this.threadId?1:0}}getActiveEventIds(){return this.activeEventId?[...

High

Trigger Reachable Command Output Exfiltration

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

dist/adapter/codex/codex-bridge.jsView on unpkg · L1

dist/core/proxy/hermes-profile-relay.pyView file

•path = dist/core/proxy/hermes-profile-relay.py kind = build_helper sizeBytes = 21259 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

dist/core/proxy/hermes-profile-relay.pyView on unpkg

Source & flagged code

11 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/install-guardian.mjs

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

openclaw-plugin/index.jsView file

3107// src/openclaw/gateway-provider.ts L3108: import { execFile } from "node:child_process"; L3109: import { readFile, writeFile as writeFile2, mkdir as mkdir2, unlink as unlink2 } from "node:fs/promises";

High

Child Process

Package source references child process execution.

openclaw-plugin/index.jsView on unpkg · L3107

3058} L3059: function buildExecApprovalResolveArgv(params) { L3060: const cliArgvPrefix = params.cliArgvPrefix && params.cliArgvPrefix.length > 0 ? params.cliArgvPrefix : resolveOpenClawCliArgvPrefix();

High

Shell

Package source references shell execution.

openclaw-plugin/index.js

Findings

1 Critical8 High4 Medium6 Low

CriticalPrevious Version Dangerous Deltadist/adapter/claude/claude-adapter.js

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processopenclaw-plugin/index.js

HighShellopenclaw-plugin/index.js

HighSame File Env Network Executiondist/adapter/openhuman/openhuman-adapter.js

HighCommand Output Exfiltrationdist/core/provider-quota/kiro.js

HighSandbox Evasion Gated Capabilitydist/core/installer/npm-registry.js

HighCross File Remote Execution Contextdist/adapter/claude/claude-adapter.js

HighTrigger Reachable Command Output Exfiltrationdist/adapter/codex/codex-bridge.js

MediumNetwork

MediumEnvironment Vars

Affected versions and remediation

This report applies to grix-connector@3.26.0.

See all versions of grix-connector for other recorded verdicts.

Review the evidence and your use of grix-connector@3.26.0 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: 2026-08-10 07:18:01.423Z (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