registry / AI-agent capability abuse / pennyrouter / 0.2.32

pennyrouter@0.2.32

Install and manage PennyRouter local coding-agent integrations.

AI Security Review

scanned 11d ago · by lpm-firewall-ai

The source implements user-invoked, first-party coding-agent integration setup, which warrants a warning under the agent configuration policy. The manifest has no install hook, and inspection found no concrete malicious behavior or unrelated credential recipient.

Static reason

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

Trigger

A user runs the pennyrouter install command.

Impact

These are concrete agent configuration and credential-forwarding capabilities, but the inspected behavior is package-aligned and did not establish an attack.

Mechanism

The CLI writes PennyRouter-owned agent configuration and registers its MCP command; its statusline sends a session identifier and API key to the selected gateway.

AI rationale

Evidence

~/.claude/settings.json~/.claude.json~/.codex/config.toml

Network endpoints1

api.pennyrouter.com

Decision evidence

public snapshot

AI called this Suspicious at 87.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The manifest exposes a CLI and has no install lifecycle hook, so these changes require a user to run the command.
The runtime installer reads Claude Code settings and configures a statusline script.
That script sends a session identifier and API key to the configured gateway’s statusline endpoint.
The package registers its own MCP server for Claude Code and Codex; the inspected registration logic preserves entries it did not create.
The flagged session-path locations identify local Claude Code and Codex session directories; the cited test locations show test fixture setup.

Evidence against

The package describes itself as a coding-agent integration manager, consistent with its configuration behavior.
The statusline request targets the PennyRouter gateway, and no unrelated credential recipient was identified.
No source evidence showed unconsented install-hook mutation or a concrete attack chain.

Behavioral surface

Source

ChildProcessCryptoEnvironmentVarsFilesystem

Source & flagged code

6 flagged

src/launch.test.jsView file

28async function main() { L29: if (!(["darwin", "linux"].includes(process.platform))) { L30: console.log("launch.test.js: skipped on unsupported platform"); ... L34: const root = fs.mkdtempSync(path.join(os.tmpdir(), "pr-launch-test-")); L35: const previous = Object.fromEntries(ENV_KEYS.map((key) => [key, process.env[key]])); L36: const home = path.join(root, "home"); ... L45: fs.writeFileSync(path.join(home, ".codex", "config.toml"), nativeCodex); L46: fs.writeFileSync(path.join(home, ".zshrc"), "# user setting\n"); L47: ... L74: apiKey: "pr-super-secret", L75: gatewayBaseUrl: "https://gateway.example/", L76: subscriptionAuth: true,

Medium

Install Persistence

Source writes installer persistence such as shell profile or service configuration.

src/launch.test.jsView on unpkg · L28

src/harnesses/codegpt.jsView file

1import { spawnSync } from "node:child_process"; L2: ... L56: console.log(" Open CodeGPT settings and remove or replace the PennyRouter provider:"); L57: console.log(" - Base URL: https://api.pennyrouter.com/v1"); L58: console.log(" - API Key: pr-..."); ... L64: const commands = L65: process.platform === "darwin" L66: ? [["pbcopy"]] L67: : process.platform === "win32"

High

Sandbox Evasion Gated Capability

Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.

src/harnesses/codegpt.jsView on unpkg · L1

src/statusline.jsView file

•Runtime or CLI code links a write operation to an explicit external AI-agent control path: code (installed by \`pennyrouter install\`). // prints: [wrapped original statusline (all lines) |] penny segment (penny / penny (sub) // / penny (byok) / penny off). the wrapped statusline's first line gets the penny suffix; // any further lines (e.g. a context bar) pass through verbatim. // never blocks on the network: reads a per-session cache and refreshes it detached. import { readfilesync, mkdirsync, writefilesync, statsync } from "node:fs"; import { spawn, spawnsync } from "node:child_process"; import { join, basename } from "node:path"; import os from "node:os"; const statedir = join( process.env.xdg_state_home || join(os.homedir(), ".local", "state"), "pennyrouter", ); const cachedir =

High

External Ai Agent Control Surface Mutation

Runtime or CLI source writes behavior-bearing configuration into a user or project AI-agent control surface.

src/statusline.jsView on unpkg

src/launch.jsView file

•Manifest-trigger-reachable source links an external AI-agent control path to a behavior-bearing write operation. import { spawn, spawnsync } from "node:child_process"; import { chmod, cp, mkdir, rm, stat, unlink, writefile, } from "node:fs/promises"; import { constants as osconstants } from "node:os"; import { fileurltopath } from "node:url"; import { basename, dirname, join, resolve } from "node:path"; import { configureclaudecodestatusline } from "./statusline.js"; import { upgradeclaudecodemodelvalue, [redacted] } from "./harnesses/claude-code.js"; import { atomicwrite, homedir, readjsonfile, readtext } from "./state.js"; const config_version = 1; const path_start = "# >>> pennyrouter path >>>"; const path_end = "# <<< pennyrouter path <<<"; const codex_profile_marker =

High

Trigger Reachable External Ai Agent Control Surface Mutation

Manifest-trigger-reachable source writes behavior-bearing configuration into a user or project AI-agent control surface.

src/launch.jsView on unpkg

src/mcp-register.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 3

High

Semantic Analysis Limited

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

src/mcp-register.js#virtual:normalized:round1View on unpkg

src/cli.jsView file

•matchType = previous_version_dangerous_delta matchedPackage = pennyrouter@0.3.17 matchedIdentity = npm:cGVubnlyb3V0ZXI:0.3.17 similarity = 0.613 summary = stored previous version shares package body but lacks this dangerous source file

High

Previous Version Dangerous Delta

This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.

src/cli.jsView on unpkg

Source & flagged code

6 flagged · loading source

src/launch.test.jsView file

Medium

Install Persistence

Source writes installer persistence such as shell profile or service configuration.

src/launch.test.jsView on unpkg · L28

src/harnesses/codegpt.jsView file

High

Sandbox Evasion Gated Capability

Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.

Findings

5 High4 Medium4 Low

HighSandbox Evasion Gated Capabilitysrc/harnesses/codegpt.js

HighExternal Ai Agent Control Surface Mutationsrc/statusline.js

HighTrigger Reachable External Ai Agent Control Surface Mutationsrc/launch.js

HighSemantic Analysis Limitedsrc/mcp-register.js#virtual:normalized:round1

HighPrevious Version Dangerous Deltasrc/cli.js

MediumNetwork

MediumEnvironment Vars

MediumInstall Persistencesrc/launch.test.js

MediumStructural Risk Force Deep Review

LowScripts Present

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

Affected versions and remediation

This report applies to pennyrouter@0.2.32.

See all versions of pennyrouter for other recorded verdicts.

Review the evidence and your use of pennyrouter@0.2.32 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: 2026-10-11 19:01:59.145Z (UTC).

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