aicodeman@1.33.1

Mission control for AI coding agents - run 20 autonomous agents with real-time monitoring and session persistence

AI Security Review

scanned 26d ago · by lpm-firewall-ai

The source supports a guarded, first-party agent-extension lifecycle risk rather than a confirmed malicious payload or credential-exfiltration path. The automatic detached server start adds concern but does not by itself prove malicious intent.

Static reason

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

Trigger

Installing the package triggers the server start; creating a Codeman-managed Claude session triggers workspace-hook setup.

Impact

This is a guarded agent-extension lifecycle capability that can alter a selected workspace's Claude configuration and process future agent events. No confirmed external data theft or foreign broad control-surface mutation was established.

Mechanism

The session path writes a Claude settings hook block that sends hook-event data to the configured Codeman API. The install script launches the package web entrypoint as a detached child process.

AI rationale

Evidence

dist/hooks-config.jsscripts/postinstall.js.claude/settings.local.json.git/hooks/pre-commit

Decision evidence

public snapshot

AI called this Suspicious at 84.0% confidence as Dangerous Capability with low false-positive risk.

Evidence for warning

The package runs a post-install script automatically.
The install script starts a detached Codeman web-server process.
When a Codeman session is created, workspace hooks are enabled by default and can write Claude settings.
The generated Claude hooks forward tool-event input to the Codeman API endpoint configured for that session.
The hook writer has safeguards for its target and the skill installer refuses symlinked or foreign skill directories.

Evidence against

No self-dependency was declared.
No static external exfiltration endpoint was identified in the inspected behavior.
The install-time server is described and configured for loopback use.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShellWebSocket

Supply chain

HighEntropyStringsMinified

Source & flagged code

29 flagged

package.jsonView file

•scripts.postinstall = node scripts/postinstall.js

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.js

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

dist/ai-checker-base.jsView file

Lines 4-44javascript

4 * Provides shared functionality for AI-powered checkers that spawn fresh Claude CLI

5 * sessions to analyze terminal output. This base class handles:

6 * - Mux session spawning and cleanup

7 * - Temp file management for output capture

8 * - Polling for completion markers

9 * - Cooldown management

10 * - Error handling and consecutive error tracking

11 * - Event emission for state changes

12 *

13 * Subclasses implement:

14 * - `muxNamePrefix`: Prefix for mux session names (e.g., 'codeman-aicheck-')

15 * - `doneMarker`: Completion marker in output file (e.g., '__AICHECK_DONE__')

16 * - `tempFilePrefix`: Prefix for temp files (e.g., 'codeman-aicheck')

17 * - `logPrefix`: Prefix for log messages (e.g., '[AiIdleChecker]')

18 * - `buildPrompt()`: Build the prompt from terminal buffer

19 * - `parseVerdict()`: Parse the verdict from AI output

20 * - `getPositiveVerdict()`: Return the "positive" verdict that doesn't trigger cooldown

21 * - `defaultConfig`: Default configuration values

22 *

23 * @module ai-checker-base

24 */

25import { execSync, spawn as childSpawn } from 'node:child_process';

26import { existsSync, readFileSync, unlinkSync, writeFileSync } from 'node:fs';

High

Child Process

Package source references child process execution.

dist/ai-checker-base.jsView on unpkg · L24

27import { tmpdir } from 'node:os';

28import { join } from 'node:path';

29import { EventEmitter } from 'node:events';

30import { getAugmentedPath, ANSI_ESCAPE_PATTERN_SIMPLE } from './utils/index.js';

31import { AI_CHECK_MAX_BACKOFF_MS } from './config/ai-defaults.js';

32import { getErrorMessage } from './types.js';

33// ========== Security Validation ==========

34/**

35 * Validates that a model name is safe for shell use.

36 * Model names should only contain alphanumeric characters, hyphens, underscores, and dots.

37 * Exported for the Read My Mind predictor, which reuses these spawn mechanics standalone.

38 */

39export function isValidModelName(model) {

40 if (!model || typeof model !== 'string')

41 return false;

42 // Allow: alphanumeric, hyphens, underscores, dots, slashes (for model paths like claude/opus-4.5)

43 // Max length 100 to prevent abuse

44 return /^[a-zA-Z0-9._/-]+$/.test(model) && model.length <= 100;

dist/remote-hosts.jsView file

Lines 1-27javascript

1import { existsSync, mkdirSync } from 'node:fs';

2import fs from 'node:fs/promises';

3import { join } from 'node:path';

4import { homedir } from 'node:os';

dist/web/repo-status.jsView file

Lines 23-63javascript

23 *

24 * PERFORMANCE: every git invocation is ASYNC (`execFile`), never `execFileSync`

25 * — the fetch path can spend `2 × FETCH_TIMEOUT_MS` per remote on the network,

26 * and a synchronous version froze the whole event loop (SSE

dist/service-installer.jsView file

Lines 6-46javascript

6 * now" half in daemon-control.ts. `install.sh` already does this for people who

7 * install with the one-liner; this exists for `npm i -g aicodeman` users, who

8 * otherwise have to hand-write a plist.

9 *

dist/deepseek-web-server.jsView file

Lines 147-187javascript

147 * prevent. The second caller now simply waits and reuses the first's server.

148 */

149let startLock = Promise.resolve();

150/**

151 * Start (or reuse) the background `dsh web` for `authority`.

152 *

153

dist/web/webview-egress-policy.jsView file

6* at different stages: the Zod schema (a URL being saved), the sync check on a L7: * hostname that is already an IP literal (Node's `net.connect` skips DNS for L8: * those, so a lookup hook never sees them), and the DNS lookup hook that judges ... L123: * connect to. Non-IP input is never blocked here: names are judged by L124: * `isBlockedWebviewHostname()` at save time and by their resolved addresses at L125: * connect time.

High

Cloud Metadata Access

Source reaches cloud instance metadata or link-local credential endpoints.

dist/web/webview-egress-policy.jsView on unpkg · L6

scripts/postinstall.jsView file

23join(home, '.local/bin/claude'), L24: join(home, '.claude/local/claude'), L25: '/usr/local/bin/claude', ... L256: L257: const { mkdirSync, copyFileSync } = await import('fs'); L258: mkdirSync(vendorDir, { recursive: true }); L259: copyFileSync(join(xtermDir, 'css', 'xterm.css'), join(vendorDir, 'xterm.css')); L260: ... L269: // Fallback: copy unminified L270: copyFileSync(join(xtermDir, 'lib', 'xterm.js'), join(vendorDir, 'xterm.min.js')); L271: copyFileSync(join(fitDir, 'lib', 'addon-fit.js'), join(vendorDir, 'xterm-addon-fit.min.js')); L272: copyFileSync(join(serializeDir, 'lib', 'addon-serialize.js'), join(vendorDir, 'xterm-addon-serialize.min.js'));

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.jsView on unpkg · L23

7Cross-file remote execution chain: scripts/postinstall.js spawns dist/web/public/vendor/xterm-addon-webgl.min.js; helper contains network access plus dynamic code execution. L7: L8: import { execSync, spawn } from 'child_process'; L9: import { existsSync } from 'fs'; ... L20: // Claude CLI search paths (must match src/session.ts) L21: const home = homedir(); L22: const CLAUDE_SEARCH_PATHS = [ ... L33: L34: const useColor = process.stdout.isTTY && !process.env.NO_COLOR; L35: ... L83: const ok = await new Promise((resolve) => { L84: const conn = net.createConnection({ port, host: '127.0.0.1' }); L85: conn.once('connect', () => { conn.destroy(); resolve(true); });

High

Cross File Remote Execution Context

Source spawns a local helper that also contains network and dynamic execution context; review data flow before blocking.

scripts/postinstall.jsView on unpkg · L7

dist/web/public/gesture/wasm/vision_wasm_nosimd_internal.wasmView file

•path = dist/web/public/gesture/wasm/vision_wasm_nosimd_internal.wasm kind = wasm_module sizeBytes = 9448638 magicHex = [redacted]

Medium

Ships Wasm Module

Package ships WebAssembly modules.

dist/web/public/gesture/wasm/vision_wasm_nosimd_internal.wasmView on unpkg

skills/codeman/preamble.shView file

•path = skills/codeman/preamble.sh kind = build_helper sizeBytes = 20196 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

skills/codeman/preamble.shView on unpkg

dist/web/public/panels-ui.67646701.js.gzView file

•path = dist/web/public/panels-ui.67646701.js.gz kind = high_entropy_blob sizeBytes = 30301 magicHex = [redacted]

High

Ships High Entropy Blob

Package ships high-entropy non-source blobs.

dist/web/public/panels-ui.67646701.js.gzView on unpkg

•path = dist/web/public/panels-ui.67646701.js.gz kind = compressed_blob sizeBytes = 30301 magicHex = [redacted]

Medium

Ships Compressed Blob

Package ships compressed or archive-like blobs.

dist/web/public/panels-ui.67646701.js.gzView on unpkg

dist/agent-case-marker.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 9

High

Semantic Analysis Limited

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

dist/agent-case-marker.js#virtual:normalized:round1View on unpkg

dist/web/routes/system-routes.jsView file

•matchType = token_shingles matchedPackage = aicodeman@1.24.6 matchedPath = dist/web/routes/system-routes.js matchedIdentity = npm:YWljb2RlbWFu:1.24.6 similarity = 1.000 shingleOverlap = 48 summary = source token shingles overlapped finalized malicious source

High