registry / AI-agent control-surface risk / pgautopilot

pgautopilot

PGAutoPilot — an MCP server that lets AI assistants safely explore, query, and manage any PostgreSQL database. Zero setup, auto-discovery, production-safe.

This report shows the evidence for pgautopilot@2.4.0. 3 of 6 known versions are blocked or flagged. See all versions.

AI Security Review

scanned 1mo ago · by lpm-firewall-ai

This is unconsented postinstall mutation of foreign AI-agent control surfaces and project skill metadata. The automatic lifecycle behavior is sufficient for a publish block even though no data exfiltration was found in the inspected code.

Static reason

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

Trigger

Installing the package through npm runs postinstall.

Impact

A package install can cause AI clients in the consumer project or user profile to launch pgautopilot without an explicit setup command.

Mechanism

Automatic AI-agent configuration and skill-lock mutation.

Policy narrative

On npm installation, package.json invokes scripts/postinstall.mjs. That script adds pgautopilot MCP launch entries for many AI tools, writes a Codex configuration, and changes an existing Claude Desktop configuration in the user profile. It then executes merge-skills-lock.mjs against the consumer project, which modifies an existing skill lock or creates one. These changes occur without an explicit user setup command and create persistent AI-agent launch and capability configuration.

AI rationale

Evidence

package.jsonscripts/postinstall.mjsscripts/merge-skills-lock.mjsconfig/cursor/mcp.jsonconfig/vscode/mcp.jsonconfig/codex/config.tomlconfig/claude/mcp.jsonskills-lock.json.skill-lock.jsonskills.lockskills-lock.yaml

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 runs a postinstall script automatically during npm installation.
The postinstall script registers pgautopilot as an MCP command for many AI editor clients.
It rewrites or creates project configuration files and also modifies an existing Claude Desktop configuration in the user's home directory.
It automatically launches a second script that adds package skills to existing or new project skill lock files.

Evidence against

No runtime self-dependency is declared.
The inspected install code does not send harvested data to a remote endpoint.

Behavioral surface

Source

ChildProcessCryptoEnvironmentVarsEvalFilesystemNetworkShell

Supply chain

HighEntropyStringsPossibleObfuscationUrlStrings

ManifestNo manifest risk signals triggered.

scanned 32 file(s), 2.12 MB of source, external domains: github.com, json-schema.org, opencode.ai, raw.githubusercontent.com, www.postgresql.org

Source & flagged code

12 flagged

package.jsonView file

Lines 1-31json

2 "name": "pgautopilot",

3 "version": "2.4.0",

4 "type": "module",

5 "description": "PGAutoPilot — an MCP server that lets AI assistants safely explore, query, and manage any PostgreSQL database. Zero setup, auto-discovery, production-safe.",

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

7 "bin": {

8 "pgautopilot": "dist/pgautopilot.bundle.cjs"

9 },

10 "scripts": {

11 "postinstall": "node scripts/postinstall.mjs",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L11

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg · L11

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

13 "postbuild": "node scripts/bundle.mjs",

14 "bundle": "node scripts/bundle.mjs",

15 "sign": "node scripts/sign.mjs",

16 "sign:gpg": "node scripts/sign.mjs --gpg",

17 "verify": "node scripts/verify.mjs",

18 "verify:gpg": "node scripts/verify.mjs --gpg",

19 "security": "npm audit --omit=dev --audit-level=high",

20 "start": "node dist/index.js",

21 "start:bundle": "node dist/pgautopilot.bundle.cjs",

22 "dev": "tsx src/index.ts",

23 "typecheck": "tsc -p tsconfig.json",

24 "lint": "tsc -p tsconfig.json && eslint src tests",

25 "test": "vitest run --dir tests",

26 "format": "prettier --write \"src/**/*.ts\" \"tests/**/*.ts\"",

27 "prepublishOnly": "npm run build"

28 },

29 "keywords": [

30 "mcp",

31 "mcp-server",

dist/pgautopilot.bundle.cjsView file

47550// dist/toolHandlers.js L47551: var import_node_child_process = require("node:child_process"); L47552: var import_fs3 = require("fs");

High

Child Process

Package source references child process execution.

dist/pgautopilot.bundle.cjsView on unpkg · L47550

1232// validation function arguments L1233: data: new codegen_1.Name("data"), L1234: // data passed to validation function ... L2259: id = normalizeId(id); L2260: return resolver.resolve(baseId, id); L2261: } ... L3122: for (i = 0; i < input2.length; i++) { L3123: code = input2[i].charCodeAt(0); L3124: if (code === 48) { ... L13346: if (!isUTC) { L13347: result.setTime(result.getTime() + result.getTimezoneOffset() * 6e4); L13348: }

High

Command Output Exfiltration

Source combines command execution, command-output handling, and outbound requests; review data flow before blocking.

dist/pgautopilot.bundle.cjsView on unpkg · L1232

•matchType = normalized_sha256 matchedPackage = pgautopilot@2.2.2 matchedPath = dist/pgautopilot.bundle.cjs matchedIdentity = npm:cGdhdXRvcGlsb3Q:2.2.2 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/pgautopilot.bundle.cjsView on unpkg

scripts/postinstall.mjsView file

Lines 1-21javascript

1import { existsSync, mkdirSync, readFileSync, writeFileSync } from "fs";

2import { resolve, dirname, join } from "path";

Critical

Ai Agent Control Hijack

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

scripts/postinstall.mjsView on unpkg · L1

3import { homedir, platform } from "os";

scripts/bundle.mjsView file

Lines 1-25javascript

1import * as esbuild from "esbuild";

2import { readFileSync, writeFileSync, existsSync, mkdirSync } from "fs";

3import { resolve, dirname } from "path";

4import { fileURLToPath }

uninstall.ps1View file

•path = uninstall.ps1 kind = build_helper sizeBytes = 1560 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

uninstall.ps1View on unpkg

src/toolHandlers.ts#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.

src/toolHandlers.ts#virtual:normalized:round1View on unpkg

install.shView file

•matchType = normalized_sha256 matchedPackage = pgautopilot@2.2.2 matchedPath = install.sh matchedIdentity = npm:cGdhdXRvcGlsb3Q:2.2.2 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.

install.shView on unpkg

Source & flagged code

12 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/postinstall.mjs

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.mjs

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

dist/pgautopilot.bundle.cjsView file

47550// dist/toolHandlers.js L47551: var import_node_child_process = require("node:child_process"); L47552: var import_fs3 = require("fs");

High

Child Process

Package source references child process execution.

dist/pgautopilot.bundle.cjsView on unpkg · L47550

Findings

1 Critical8 High5 Medium7 Low

CriticalAi Agent Control Hijackscripts/postinstall.mjs

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processdist/pgautopilot.bundle.cjs

HighShell

HighCommand Output Exfiltrationdist/pgautopilot.bundle.cjs

HighCross File Remote Execution Contextscripts/bundle.mjs

HighSemantic Analysis Limitedsrc/toolHandlers.ts#virtual:normalized:round1

HighKnown Malware Source Similaritydist/pgautopilot.bundle.cjs

HighKnown Malware Source Similarityinstall.sh

MediumAmbiguous Install Lifecycle Scriptpackage.json

MediumNetwork

MediumEnvironment Vars

Affected versions and remediation

This report applies to pgautopilot@2.4.0.

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

Avoid installing pgautopilot@2.4.0. 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: 2026-09-13 22:06:43.397Z (UTC).

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

Versions

6 known versions

2.4.0latestshownMaliciousSoft block: AI-agent control surfaceBlockPublishedSep 2, 2026, 5:23 PM 2.2.2MaliciousSoft block: AI-agent control surfaceBlockPublishedSep 1, 2026, 4:20 PM 2.1.3Suspicious—WarnUnder ReviewAug 12, 2026, 7:28 PM 2.1.2Suspicious—WarnUnder ReviewAug 12, 2026, 5:40 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

Runtime surface

Behavioral surface

Related packages

Other AI-agent control-surface risk packages

Source & flagged code