registry / AI-agent control-surface risk / @maxiaochao/pi-toolkit / 0.10.14

@maxiaochao/pi-toolkit@0.10.14

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

AI Security Review

scanned 14d ago · by lpm-firewall-ai

The package's automatic postinstall hook replaces a user-wide AI-agent instruction file without an explicit consent step. This meets the policy for blocking unconsented install-time mutation of a broad agent control surface.

Static reason

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

Trigger

Installing the npm package triggers its postinstall hook.

Impact

The package can replace instructions that affect future Pi agent behavior and add a user-level command and shell completion.

Mechanism

The hook writes the bundled instructions to ~/.pi/agent/AGENTS.md when that file is missing or differs, then copies the pi-worktree command and completion into ~/.local directories.

Policy narrative

Installing the package automatically runs a hook that replaces the global Pi agent instruction file when it differs from the bundled version. The same hook installs a command and shell completion in user-wide locations. This is an unconsented install-time change to an AI-agent control surface.

AI rationale

Evidence

package.jsonscripts/install-agents.mjspackages/worktree/scripts/install.mjsglobal/AGENTS.md~/.pi/agent/AGENTS.md~/.local/bin/pi-worktree~/.local/share/bash-completion/completions/pi-worktree

Decision evidence

public snapshot

AI identified this as a dangerous AI-agent capability at 97.0% confidence with low false-positive risk.

Evidence for policy risk

package.json runs scripts/install-agents.mjs automatically after installation.
The hook targets the user-wide ~/.pi/agent/AGENTS.md and replaces it with the package's bundled file when contents differ.
The hook also installs the pi-worktree command and shell completion into user-wide ~/.local paths.
The bundled AGENTS.md supplies behavioral instructions for an AI coding agent.

Evidence against

The bundled instructions are described as general coding guidance, but the hook still replaces the global file without an explicit consent step.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShellWebSocket

Supply chain

HighEntropyStringsUrlStrings

Manifest

NoLicense

scanned 81 file(s), 682 KB of source, external domains: 127.0.0.1, example.com, github.com

Source & flagged code

14 flagged

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

packages/worktree/tests/extension.test.mjsView file

57L58: const extension = (await import(`${extensionUrl}?test=${Math.random()}`)).default; L59: extension(pi);

Medium

Dynamic Require

Package source references dynamic require/import behavior.

packages/worktree/tests/extension.test.mjsView on unpkg · L57

scripts/install-agents.mjsView file

1// postinstall hook: sync the bundled global AGENTS.md and seed the default theme L2: // without overriding a theme the user has already selected. L3: import { readFileSync, writeFileSync, mkdirSync, existsSync } from "node:fs"; L4: import { homedir } from "node:os"; ... L11: const agentDir = join(homedir(), ".pi", "agent"); L12: const bundled = readFileSync(join(dirname(fileURLToPath(import.meta.url)), "..", "global", "AGENTS.md"), "utf8"); L13: const agentsTarget = join(agentDir, "AGENTS.md"); L14: const settingsTarget = join(agentDir, "settings.json"); ... L16: if (!existsSync(agentsTarget) || readFileSync(agentsTarget, "utf8") !== bundled) { L17: mkdirSync(dirname(agentsTarget), { recursive: true }); L18: writeFileSync(agentsTarget, bundled, "utf8"); L19: console.log("[pi-toolkit]

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/install-agents.mjsView on unpkg · L1

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

packages/worktree/bin/pi-worktree-completion.bashView file

•path = packages/worktree/bin/pi-worktree-completion.bash kind = build_helper sizeBytes = 1384 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

packages/worktree/bin/pi-worktree-completion.bashView on unpkg

packages/worktree/tests/run-tests.shView file

•path = packages/worktree/tests/run-tests.sh kind = payload_in_excluded_dir sizeBytes = 4742 magicHex = [redacted]

High

Payload In Excluded Dir

Package hides binary, compressed, or executable-looking payloads in test/fixture/hidden paths.

packages/worktree/tests/run-tests.shView on unpkg

packages/codex-edit/tests/generate-test-report.mjs#virtual:string-array: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.

packages/codex-edit/tests/generate-test-report.mjs#virtual:string-array:round1View on unpkg

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

•matchType = normalized_sha256 matchedPackage = @maxiaochao/pi-toolkit@0.10.13 matchedPath = skills/web-browser/scripts/start.js matchedIdentity = npm:QG1heGlhb2NoYW8vcGktdG9vbGtpdA:0.10.13 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High