aims-workstation@0.1.24
AIMS 工作站：本地智能体工作站——ARM 助手本地化（下载/离线/升级）、工作台对话式定制、多工作区隔离、分级自主与审计
AI Security Review
scanned 5d ago · by lpm-firewall-aiThis is a serious default network exposure: the server listens on all interfaces while API authentication is disabled unless the user configures a token. The agent shell tool makes unauthenticated chat control a critical vulnerability, though the server must first be explicitly started.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
The user starts the server, then a network peer sends a request to POST /api/chat while no access token has been configured.
Impact
A network peer may direct deployed agents to access workspace or other readable data and receive results through the chat response. Write commands have a confirmation flow, which limits but does not remove the exposure.
Mechanism
The server binds to 0.0.0.0, its access token defaults to empty, and the authentication check allows requests when the token is unset. The chat endpoint invokes an agent loop that has a shell execution tool.
AI rationale
This is a serious default network exposure: the server listens on all interfaces while API authentication is disabled unless the user configures a token. The agent shell tool makes unauthenticated chat control a critical vulnerability, though the server must first be explicitly started.
Evidence
server.jsAIMS_HOME/workspace/<agent>
Network endpoints2
0.0.0.0:5600POST /api/chat
Decision evidence
public snapshotAI called this Suspicious at 96.0% confidence as Critical Vulnerability with medium false-positive risk.
Evidence for warning
- The server binds to 0.0.0.0 by default, making it reachable from the local network.
- The access token defaults to an empty string, and authOk allows requests when no token is configured.
- The API chat handler accepts a caller-supplied agent and message, then runs the agent loop.
- The agent exec tool passes commands to cpExec through a shell.
Evidence against
- The server starts only when the user explicitly runs the package command; there is no install hook.
- The package is described as a workstation intended for team sharing, and its README documents the LAN-accessible default.
- Write commands are subject to an L2 human-confirmation flow.
Behavioral surface
ChildProcessCryptoEnvironmentVarsEvalFilesystemNetworkShell
HighEntropyStringsMinifiedUrlStrings
NoLicense
Source & flagged code
6 flaggedserver.jsView file
6// 覆盖所有出站 fetch（ARM REST、LLM、面板代理）。本工具面向内网/本机使用。
L7: process.env.NODE_TLS_REJECT_UNAUTHORIZED = '0';
L8: const http = require('http');
L9: const fs = require('fs');
...
L23: // ── 代码与数据分离：包目录只放代码与只读资产；全部运行时状态在 AIMS_HOME（缺省 ~/.aims-workstation）──
L24: const PKG_ROOT = __dirname;
L25: const AIMS_HOME = process.env.AIMS_HOME ? path.resolve(process.env.AIMS_HOME) : path.join(os.homedir(), '.aims-workstation');
...
L81: // ro=只读白名单命令名（免确认直行）；sudo_u_l2=true 时 `sudo -u <用户>` 降权形式不再永久拒绝、按写命令走 L2 人工确认（裸 sudo 仍拒绝）
L82: // 默认清单按平台合成：POSIX 为主；Windows（cmd/PowerShell 命令与毁灭性命令）仅在 win32 追加——非 Windows 平台不引入 "format" 等词的误杀
L83: const WIN = process.platform === 'win32';
L84: const GUARD_POSIX = {
...
L127: function loadCfg() {
Critical
Credential Exfiltration
Source appears to send environment or credential material to an external endpoint.
server.jsView on unpkg · L66Trigger-reachable chain: scripts.start -> server.js
L6: // 覆盖所有出站 fetch（ARM REST、LLM、面板代理）。本工具面向内网/本机使用。
L7: process.env.NODE_TLS_REJECT_UNAUTHORIZED = '0';
L8: const http = require('http');
L9: const fs = require('fs');
...
L23: // ── 代码与数据分离：包目录只放代码与只读资产；全部运行时状态在 AIMS_HOME（缺省 ~/.aims-workstation）──
L24: const PKG_ROOT = __dirname;
L25: const AIMS_HOME = process.env.AIMS_HOME ? path.resolve(process.env.AIMS_HOME) : path.join(os.homedir(), '.aims-workstation');
...
L81: // ro=只读白名单命令名（免确认直行）；sudo_u_l2=true 时 `sudo -u <用户>` 降权形式不再永久拒绝、按写命令走 L2 人工确认（裸 sudo 仍拒绝）
L82: // 默认清单按平台合成：POSIX 为主；Windows（cmd/PowerShell 命令与毁灭性命令）仅在 win32 追加——非 Windows 平台不引入 "format" 等词的误杀
L83: const WIN = process.platform === 'win32';
L84: const GUARD_POSIX = {
...
L127: function loadCfg() {
Critical
Trigger Reachable Dangerous Capability
A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
server.jsView on unpkg · L6•matchType = previous_version_dangerous_delta
matchedPackage = aims-workstation@0.1.12
matchedIdentity = npm:YWltcy13b3Jrc3RhdGlvbg:0.1.12
similarity = 0.833
summary = stored previous version shares package body but lacks this dangerous source file
Critical
Previous Version Dangerous Delta
This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.
server.jsView on unpkg•Manifest-reachable source captures an API credential, sends it to a fixed unofficial gateway, and persists that redirection in agent or shell configuration.
server.js:
if (s.includes(oldP)) fs.writeFileSync(reg, s.split(oldP).join(newP));
function saveCfg() { fs.writeFileSync(CFG_FILE, JSON.stringify(cfg, null, 2)); }
// LLM：OpenAI 兼容端点（baseUrl/模型/密钥），缺省公司内网 bifrost 网关（内网可达即可用，无需配置）
const llmBaseUrl = () => cfg.llm_base_url || env.LLM_BASE_URL || LLM_DEFAULT_BASE;
return req.headers.authorization === `Bearer ${accessToken()}`;
const base = llmBaseUrl(), key = llmApiKey(), name = llmModelName();
// 官方模型条目为模板（compat/thinkingLevelMap 等字段不自行猜测）；任意模型名按 glm-5.3 克隆；条目 baseUrl 一并改写
const entries = zaiCodingCnProvider().getModels().map((m) => ({ ...m, provider: 'platform-llm', baseUrl: base }));
High
Credential Redirect Persistence
Manifest-reachable source sends a prompted API credential to a fixed unofficial gateway and persists the redirection.
server.jsView on unpkg6Trigger-reachable credential exfiltration chain: scripts.start -> server.js
L6: // 覆盖所有出站 fetch（ARM REST、LLM、面板代理）。本工具面向内网/本机使用。
L7: process.env.NODE_TLS_REJECT_UNAUTHORIZED = '0';
L8: const http = require('http');
L9: const fs = require('fs');
...
L23: // ── 代码与数据分离：包目录只放代码与只读资产；全部运行时状态在 AIMS_HOME（缺省 ~/.aims-workstation）──
L24: const PKG_ROOT = __dirname;
L25: const AIMS_HOME = process.env.AIMS_HOME ? path.resolve(process.env.AIMS_HOME) : path.join(os.homedir(), '.aims-workstation');
...
L81: // ro=只读白名单命令名（免确认直行）；sudo_u_l2=true 时 `sudo -u <用户>` 降权形式不再永久拒绝、按写命令走 L2 人工确认（裸 sudo 仍拒绝）
L82: // 默认清单按平台合成：POSIX 为主；Windows（cmd/PowerShell 命令与毁灭性命令）仅在 win32 追加——非 Windows 平台不引入 "format" 等词的误杀
L83: const WIN = process.platform === 'win32';
L84: const GUARD_POSIX = {
...
L127: function loadCfg() {
High
Trigger Reachable Credential Exfiltration
A manifest entrypoint or package-local install chain reaches credential exfiltration behavior.
server.jsView on unpkg · L6scripts/publish-npm.shView file
•path = scripts/publish-npm.sh
kind = build_helper
sizeBytes = 2409
magicHex = [redacted]
Medium
Ships Build Helper
Package ships non-JavaScript build or shell helper files.
scripts/publish-npm.shView on unpkgFindings
3 Critical2 High4 Medium6 Low
CriticalCredential Exfiltrationserver.js
CriticalTrigger Reachable Dangerous Capabilityserver.js
CriticalPrevious Version Dangerous Deltaserver.js
HighCredential Redirect Persistenceserver.js
HighTrigger Reachable Credential Exfiltrationserver.js
MediumNetwork
MediumEnvironment Vars
MediumShips Build Helperscripts/publish-npm.sh
MediumStructural Risk Force Deep Review
LowScripts Present
LowEval
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
LowNo License
Affected versions and remediation
This report applies to aims-workstation@0.1.24.
See all versions of aims-workstation for other recorded verdicts.
- Review the evidence and your use of aims-workstation@0.1.24 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: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.