registry / AI-agent control-surface risk / @maxiaochao/pi-toolkit / 0.8.2

@maxiaochao/pi-toolkit@0.8.2

Personal pi coding agent toolkit: /todos, /btw, /cache_export extensions, browser automation, and bundled theme/skills.

AI Security Review

scanned 11d ago · by lpm-firewall-ai

The postinstall hook makes an unconsented write to a global AI-agent control surface by replacing its instructions with package-controlled content. This supports blocking under the agent control hijack policy.

Static reason

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

Trigger

Installing the package when ~/.pi/agent/AGENTS.md is absent or differs from the bundled file.

Impact

This changes instructions used by the user's AI agent across projects.

Mechanism

The hook reads the bundled AGENTS.md and writes it to the global agent directory without requesting consent.

Policy narrative

Package installation automatically runs a hook that synchronizes a package-supplied AGENTS.md into the user's global Pi agent directory. When the existing instructions differ, the hook overwrites them without a consent check, changing the agent's behavior across projects.

AI rationale

Evidence

package.jsonscripts/install-agents.mjsglobal/AGENTS.md~/.pi/agent/AGENTS.md

Decision evidence

public snapshot

AI identified this as a dangerous AI-agent capability at 94.0% confidence with low false-positive risk.

Evidence for policy risk

The package runs scripts/install-agents.mjs automatically through its postinstall hook.
The hook targets the user's global ~/.pi/agent/AGENTS.md and loads the package's bundled instructions.
If the existing file differs, the hook replaces it without a consent check.
The bundled file supplies behavioral instructions for the user's AI agent.

Evidence against

The hook preserves an identical AGENTS.md and initializes or fills in only a missing theme setting in settings.json.
The inspected browser process execution is in an end-to-end test, and the inspected worktree CLI performs user-invoked Git operations.

Behavioral surface

Source

ChildProcessEnvironmentVarsFilesystemNetworkShellWebSocket

Supply chain

HighEntropyStringsUrlStrings

Manifest

NoLicense

scanned 24 file(s), 233 KB of source, external domains: 127.0.0.1, example.com

Source & flagged code

10 flagged

package.jsonView file

Lines 34-60json

34 "./themes"

35 ]

36 },

37 "bin": {

38 "pi-worktree": "bin/pi-worktree"

39 },

40 "peerDependencies": {

41 "@earendil-works/pi-ai": "*",

42 "@earendil-works/pi-coding-agent": "*",

43 "@earendil-works/pi-tui": "*",

44 "typebox": "*"

45 },

46 "dependencies": {

47 "ws": "^8.19.0"

48 },

49 "scripts": {

50 "test": "cd extensions/cache-export && node --experimental-strip-types tests/run-tests.mjs && node --experimental-strip-types ../tests/btw-history.mjs && node tests/e2e-browser.mjs",

51 "load-test": "pi -ne -e ./extensions/todo.ts && pi -ne -e ./extensions/cache-export/index.ts && pi -ne -e ./extensions/btw.ts",

52 "eval:agents": "node evaluations/agents-md/run.mjs",

53 "eval:agents:test": "node evaluations/agents-md/tests/run-tests.mjs",

54 "postinstall": "node scripts/install-agents.mjs"

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L54

55 },

56 "publishConfig": {

57 "access": "public"

58 }

59}

extensions/cache-export/tests/e2e-browser.mjsView file

Lines 1-32javascript

1// Real-browser end-to-end test for the cache dashboard.

2// Drives actual pointer events in headless Chromium via raw CDP (Node 22

3// built-ins only: fetch + WebSocket) and asserts the hover tooltip keeps

4// working across polyline break boundaries and inside segment views.

5//

6// Chrome discovery mirrors the web-browser skill (BROWSER_BIN, common Linux

7// paths, Playwright cache, WSL Windows Chrome). If no browser is found the

8// test prints SKIP and exits 0 so CI without Chrome still passes; if a

9// browser IS found, failing assertions exit 1.

skills/web-browser/scripts/start.jsView file

Lines 41-81javascript

42const HOME = process.env["HOME"] || homedir();

43const CACHE_ROOT = join(HOME, ".cache", "agent-web"

skills/chrome-cdp/scripts/cdp.mjsView file

Lines 1-32javascript

1#!/usr/bin/env node

2// cdp - lightweight Chrome DevTools Protocol CLI

3// Uses raw CDP over WebSocket, no Puppeteer dependency.

4// Requires Node 22+ (built-in WebSocket).

5//

6// Per-tab persistent daemon: page commands go through a daemon that holds

7// the CDP session open. Chrome's "Allow debugging" modal fires once per

8// daemon (= once per tab). Daemons auto-exit after 20min idle.

10import { readFileSync, writeFileSync, unlinkSync, existsSync, mkdirSync }

scripts/install-agents.mjsView file

Lines 1-21javascript

1// postinstall hook: sync the bundled global AGENTS.md and seed the default theme

2// without overriding a theme the user has already selected.

3import { readFileSync, writeFileSync, mkdirSync, existsSync, copyFileSync, chmodSync } from "node:fs";

4import { homedir, platform } from "node:os";

5import { dirname, join } from "node:path";

skills/pdlog/tools/ppmdView file

•path = skills/pdlog/tools/ppmd kind = native_binary sizeBytes = 68456 magicHex = [redacted]

Medium

Ships Native Binary

Package ships native binary artifacts.

skills/pdlog/tools/ppmdView on unpkg

bin/pi-worktree-completion.bashView file

•path = bin/pi-worktree-completion.bash kind = build_helper sizeBytes = 1368 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

bin/pi-worktree-completion.bashView on unpkg

bin/pi-worktreeView file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 1

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

bin/pi-worktreeView on unpkg

Source & flagged code

10 flagged · loading source

package.jsonView file

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

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

extensions/cache-export/tests/e2e-browser.mjsView file

12L13: import { spawn, execFileSync } from "node:child_process"; L14: import { existsSync, mkdtempSync, readdirSync, rmSync } from "node:fs";

High

Child Process

Package source references child process execution.

extensions/cache-export/tests/e2e-browser.mjsView on unpkg · L12

•matchType = normalized_sha256 matchedPackage = @maxiaochao/pi-toolkit@0.5.2 matchedPath = extensions/cache-export/tests/e2e-browser.mjs matchedIdentity = npm:QG1heGlhb2NoYW8vcGktdG9vbGtpdA:0.5.2 similarity = 1.000 summary = normalized source hash matched finalized malicious source

Findings

1 Critical7 High5 Medium5 Low

CriticalAi Agent Control Hijackscripts/install-agents.mjs

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processextensions/cache-export/tests/e2e-browser.mjs

HighShellskills/web-browser/scripts/start.js

HighSame File Env Network Executionskills/chrome-cdp/scripts/cdp.mjs

HighSemantic Analysis Limitedbin/pi-worktree

HighKnown Malware Source Similarityskills/chrome-cdp/scripts/cdp.mjs

HighKnown Malware Source Similarityextensions/cache-export/tests/e2e-browser.mjs

MediumNetwork

MediumEnvironment Vars

MediumShips Native Binaryskills/pdlog/tools/ppmd

Medium

Affected versions and remediation

This report applies to @maxiaochao/pi-toolkit@0.8.2.

See all versions of @maxiaochao/pi-toolkit for other recorded verdicts.

Avoid installing @maxiaochao/pi-toolkit@0.8.2. 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: 2026-10-11 18:10:45.251Z (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

Related packages

Other AI-agent control-surface risk packages

Behavioral surface

Source & flagged code

Related packages

Other AI-agent control-surface risk packages