Loading npm security reports…
Loading npm security reports…
@minhspark/codex-mcp-bridge@1.26.0: AI-agent control-surface risk | LPM Firewall
Soft block: AI-agent control surface
Warn by default; block when configured. Agent behavior and background worker configuration are changed without a separate user command.
AI security reviewscanned on Nov 1, 2026, 06:59 UTC · by lpm-firewall-ai The postinstall hook automatically writes agent configuration outside the package directory during global installs. The package’s integration purpose does not make those install-time control-surface changes consented.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review
Trigger
npm postinstall during a global install, unless CODEX_BRIDGE_SKIP_SETUP=1 is set.
Impact
Agent behavior and background worker configuration are changed without a separate user command.
Mechanism
The setup routine invokes installers that write MCP registrations and related configuration for Claude Code, Claude Desktop, Codex, Antigravity, and a background relay.
Policy narrative
A global npm install runs the postinstall setup automatically. That setup invokes installers which modify Claude, Codex, and Antigravity agent configuration outside the package directory and configure a background relay. The agent-control changes occur during installation without a separate user command, meeting the install-time agent control-surface rule.
Evidence
package.json src/setup.mjs scripts/install-claude-code.mjs scripts/install-codex-mcp.mjs scripts/install-antigravity-bridge.mjs
scripts/setup-background-relay.mjs
~/.claude.json (or CLAUDE_CONFIG_DIR/.claude.json)
Claude Desktop configuration file
~/.codex/config.toml
~/.gemini/config/mcp_config.json
Per-user background relay service configuration
Decision evidencepublic snapshot AI identified this as a dangerous AI-agent capability at 97.0% confidence with low false-positive risk.
Evidence for policy risk
The package runs scripts/setup.mjs from its postinstall lifecycle hook. The postinstall path runs automatically for global npm installs unless CODEX_BRIDGE_SKIP_SETUP=1 is set. Setup invokes installers that write Claude Code, Codex, and Antigravity agent configuration outside the package directory. Evidence against
Automatic setup is limited to global npm installs, and the package provides an environment variable to skip it. The configuration changes serve the package’s stated purpose of connecting coding agents through MCP. Behavioral surface
Source Child process · Crypto · Dynamic require · Environment variables · Filesystem · Network · Shell · Web socket
Supply chain High entropy strings · URL strings
Manifest No manifest risk signals triggered.
Scanned 112 files · 1.24 MB of source
7 external domains referenced
api.npmjs.org api.openai.com github.com schemas.microsoft.com vdbwkowdggtcihixowxi.supabase.co www.apple.com www.w3.org Findings1 critical · 19 high · 7 medium · 4 low Source & flagged code24 flagged Critical AI agent control hijack
Source creates an unconsented AI-agent control surface through install-time mutation or a default unauthenticated remote skill channel.
package.json View on unpkg · L23 Code excerpt · 15 lines Install-time code directly mutates a foreign AI-agent control surface:
49 "mcp.json",
50 ".codex-plugin",
51 ".mcp.json",
52 "glama.json",
⋯
87 "uninstall:agent": "node scripts/install-launch-agent.mjs --uninstall",
88 "version": "node scripts/sync-version.mjs && git add src/index.mjs src/claude-bridge.mjs src/native-relay-companion.mjs scripts/install-native-relay.mjs .codex-plugin/plugin.json"
89 },
Write operation from src/setup.mjs:
23 export function automaticSetupEnabled(env = process.env) {
24 return env.npm[redacted] === "true" && env.CODEX_BRIDGE_SKIP_SETUP !== "1";
25 }
⋯
28 const home = env.HOME || env.USERPROFILE || os.homedir();
29 const codeDirectory = env.CLAUDE_CONFIG_DIR || home;
30 const configRoot = env.XDG_CONFIG_HOME || path.join(home, ".c
High Install time lifecycle scripts
Package defines install-time lifecycle scripts.
package.json View on unpkg Medium Ambiguous install lifecycle script
Install-time lifecycle script is not statically allowlisted and needs review.
package.json View on unpkg Shared scanner record · 2 findings · 1 line scripts.postinstall = node scripts/setup.mjs --postinstall
scripts/install-background-relay.mjs View file High Child process
Package source references child process execution.
scripts/install-background-relay.mjs View on unpkg · L1 Code excerpt · 2 lines 1 import { execFileSync } from "node:child_process" ;
2 import { createHash, randomUUID } from "node:crypto" ;
src/release-snapshot.mjs#virtual:normalized:round1 View file High Shell
Package source references shell execution.
src/release-snapshot.mjs#virtual:normalized:round1 View on unpkg · L248 Code excerpt · 3 lines 248 const DIGEST_FILE_PATTERN = /^\.digests-[a-f0-9]{64}\.json$/;
249 const DIGEST_HEADER = /^\{"schema":1,"directory":("(?:[^"\\]|\\.)*")/;\nconst PRUNE_LOCK = ".prune.lock";\nconst PRUNE_STAMP = ".prune-stamp";\nconst LEASE_HEARTBEAT_MS = 5 * 60 * ...
250 }
Medium Dynamic require
Package source references dynamic require/import behavior.
src/background-relay-service.mjs View on unpkg · L297 Code excerpt · 3 lines 297 if (result.status !== "reload" ) break ;
298 run = (await import (`${result.moduleFile ? pathToFileURL(result.moduleFile).href : import.meta.url}?reload=${randomUUID()}` )).runBackgroundRelayService;
299 }
High Known malware source similarity
Source file is highly similar to a previously finalized malicious package; route for source-aware review.
scripts/install-launch-agent.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = scripts/install-launch-agent.mjs
matchedIdentity = npm:[redacted]:1.25.3
similarity = 1.000
summary = normalized source hash matched finalized malicious source
Medium Install persistence
Source writes installer persistence such as shell profile or service configuration.
scripts/install-launch-agent.mjs View on unpkg · L1 Code excerpt · 11 lines 1 import { execFileSync } from "node:child_process" ;
2
scripts/publish-repo-analytics.mjs#virtual:normalized:round1 View file High Same file environment network execution
A single source file combines environment access, network access, and code or shell execution; review context before blocking.
scripts/publish-repo-analytics.mjs#virtual:normalized:round1 View on unpkg · L1 High Command output exfiltration
Source combines command execution, command-output handling, and outbound requests; review data flow before blocking.
scripts/publish-repo-analytics.mjs#virtual:normalized:round1 View on unpkg · L1 Shared code excerpt · 2 findings · 4 lines 1 import { spawn } from "node:child_process";
2 import { mkdir, readFile, writeFile } from "node:fs/promises";
⋯
6
7 const escapeXml = value => String(value ?? "").replace(/[&<>"']/g, character => ({ '&': '&', '<': '<', '>': '>', '"": """, \""": ''' })[character]);\nconst numb...
High Semantic analysis limited
High External AI agent control surface mutation
Runtime or CLI source writes behavior-bearing configuration into a user or project AI-agent control surface.
scripts/install-claude-code.mjs View on unpkg Scanner record · 16 lines Runtime or CLI code links a write operation to an explicit external AI-agent control path:
#!/usr/bin/env node
import fs from "node:fs";
import path from "node:path";
import { randomuuid } from "node:crypto";
import { fileurltopath } from "node:url";
import { claudedesktopconfigpath, homedir } from "../src/platform.mjs";
import { checkcoderegistration, plancoderegistration, planinitialcoderegistration } from "../src/claude-code-registration.mjs";
import { readprojectpolicy, readregularconfigfile, backupregularconfigfile } from "../src/project-policy.mjs";
import { exitforversionrequest } from "../src/cli-version.mjs";
import { stableinstallpath, stablenodebin } from "../src/install-paths.mjs";
exitforversionrequest(import.meta.url);
High Known malware source similarity
Source file is highly similar to a previously finalized malicious package; route for source-aware review.
scripts/collect-repo-analytics.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = scripts/collect-repo-analytics.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/app-server-client.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/app-server-client.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/native-relay.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/native-relay.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/release-snapshot.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/release-snapshot.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/remote-codex.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/remote-codex.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/sites-connector.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/sites-connector.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/antigravity-desktop.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/antigravity-desktop.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/codex-full-access.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/codex-full-access.mjs
matchedIdentity = npm:[redacted]:1.25.3
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/publish-repo-analytics.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = scripts/publish-repo-analytics.mjs
matchedIdentity = npm:[redacted]:1.25.3
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.
src/claude-desktop-context.mjs View on unpkg Scanner record · 6 lines matchType = normalized_sha256
matchedPackage = @minhspark/codex-mcp-bridge@1.25.3
matchedPath = src/claude-desktop-context.mjs
matchedIdentity = npm:[redacted]:1.25.3
similarity = 1.000
summary = normalized source hash matched finalized malicious source
AI assessment confidence 97%
Certainty of the AI review, not severity.
Verdict details False-positive risk, Review stage, Firewall policy This finding concerns changes to an AI agent's instructions or configuration without explicit user action. It does not by itself establish malware intent. The CLI warns by default and blocks when configured for this policy.
Package metadata npm Author Bui Dang Minh
Version 1.26.0
Published Nov 1, 2026
License MIT
Dependencies 2
Integrity verified
Package size 477 KB
Files 137
Keywords mcp codex claude-code claude-desktop app-server antigravity model-context-protocol
Runtime surface package.json Entrypoints bin → antigravity-mcp-bridge, antigravity-mcp-bridge-install, bridge-projects +9
Runtime node >=22.13
Artifact 1.61 MB unpacked · 137 files 2 signatures · provenance
Install lifecycle postinstall
On this report AI security review Decision evidence Findings31 Flagged code24 Remediation Code↓
Package metadata 7 findings
Medium Network
Medium Environment variables
Medium Structural risk force deep review
import fs from "node:fs" ;
18 execFileSync ("launchctl" , ["bootout" , `${domain}/${LAUNCH_AGENT_LABEL}` ], { stdio: "pipe" });
32 const codexBin = resolveCodexBin (process.env.CODEX_EXE);
33 if (!path.isAbsolute (codexBin) || !fs.existsSync (codexBin)) {
37 const url = process.env.CODEX_APP_SERVER_URL ?? "ws://127.0.0.1:8791" ;
38 const logDir = path.join (os.homedir (), "Library" , "Logs" , "codex-mcp-bridge" );
39 fs.mkdirSync (logDir, { recursive: true });
A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.
scripts/publish-repo-analytics.mjs#virtual:normalized:round1
Scanner record · 1 line stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 2
const args = process.argv.slice(2);