Back to skill

Security audit

holiday-enough

Security checks for vulnerabilities and agentic risk

Overview

This travel-planning skill asks users to enable broad control of their normal Chrome browser, which is more access than the task safely needs.

Install only if you are comfortable giving the skill control over a Chrome instance with remote debugging enabled. A safer version should use an isolated temporary browser profile, restrict actions to Xiaohongshu pages it opens itself, remove arbitrary JavaScript/file-upload/screenshot-path endpoints, authenticate the local proxy, and shut it down automatically after use.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/cdp-proxy.mjs:313
Finding

Unauthenticated Arbitrary JavaScript Execution in Authenticated Browser Tabs

Content
View full analysis
{ const parsed = new URL(req.url, `http://localhost:${PORT}`); const pathname = parsed.pathname; const q = Object.fromEntries(parsed.searchParams); res.setHeader('Content-Type', 'application/json; charset=utf-8'); try { ``` ```js // GET /targets - list all pages if (pathname === '/targets') { const resp = await sendCDP('Target.getTargets'); const pages = resp.result.targetInfos.filter(t => t.type === 'page'); res.end(JSON.stringify(pages, null, 2)); } ``` ```js // POST /eval?target=xxx - execute JavaScript else if (pathname === '/eval') { const sid = await ensureSession(q.target); const body = await readBody(req); const expr = body || q.expr || 'document.title'; const resp = await sendCDP('Runtime.evaluate', { expression: expr, returnByValue: true, awaitPromise: true, }, sid); if (resp.result?.result?.value !== undefined) { res.end(JSON.stringify({ value: resp.result.result.value })); } else if (resp.result?.exceptionDetails) { res.statusCode = 400; res.end(JSON.stringify({ error: resp.result.exceptionDetails.text })); } else { res.end(JSON.stringify(resp.result)); } } ``` ```js server.listen(PORT, '127.0.0.1', () => { console.log(`[CDP Proxy] Running at http://localhost:${PORT}`); connect().catch(e => console.error( '[CDP Proxy] Initial connection failed:', e.message, '(will retry on first request)' )); }); ``` ### Technical Analysis The proxy exposes Chrome DevTools Protocol operations over an HTTP service bound to `127.0.0.1`. Binding to loopback reduces remote network exposure, but it does not provide authentication or authorization. Any process opera ...[truncated 2632 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/cdp-proxy.mjs:484
Finding

Arbitrary Filesystem Write Through Screenshot Output Path

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/check-deps.mjs:92
Finding

Detached Browser-Control Proxy Persists Beyond the Skill Run

Content
View full analysis
setTimeout(r, 2000)); ``` ### Technical Analysis The dependency-check script starts the CDP proxy as a detached child and calls `unref()`. This allows the proxy to continue running independently after the dependency checker and Skill task terminate. There is no corresponding shutdown operation, parent-death handling, process identifier management, ownership validation, maximum lifetime, or inactivity timeout. The proxy therefore remains available until manually terminated, the system shuts down, or another external event ends it. This is not installation as an operating-system startup service or scheduled task, so it does not constitute durable reboot-level persistence. However, it unnecessarily extends a highly privileged, unauthenticated browser-control service beyond the task that required it. The behavior exceeds minimum privilege because the proxy is needed only during active retrieval of travel information. ### Attack Path 1. The Skill runs `scripts/check-deps.mjs` as instructed by `SKILL.md`. 2. If no ready proxy is detected, `startProxyDetached()` launches `cdp-proxy.mjs`. 3. The child detaches and is unreferenced, allowing the dependency-check process t ...[truncated 1050 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (18)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 29)May include surrounding context.

md
用户确认后重新运行 `check-deps.mjs` 验证连接。

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill directs the agent to execute local shell commands and interact with a localhost browser-control service that can open pages, evaluate arbitrary DOM JavaScript, scroll, and close tabs. This is effectively scripted browser control with code execution semantics against the user's browsing context, enabling access to authenticated content and potentially unrelated local/browser data.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This file implements a local HTTP-to-CDP proxy that can discover a Chrome remote-debugging port and then control the user's regular browser, tabs, navigation, JavaScript execution, clicks, screenshots, and file inputs. That capability is far broader than the declared travel-duration purpose, and in this context strongly indicates hidden general-purpose browser takeover functionality rather than legitimate trip-length estimation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The exposed endpoints allow arbitrary browser automation, including navigation, arbitrary JavaScript execution via /eval, DOM interaction, realistic clicks, file-input manipulation, scrolling, and screenshots. In a skill whose stated purpose is evaluating whether vacation days are sufficient, these capabilities are unjustified and enable credential theft, session abuse, data exfiltration, and unauthorized actions in sites already open in the user's browser.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file's behavior is materially inconsistent with the stated purpose of evaluating whether vacation length is sufficient for a destination. By performing environment checks for Chrome debugging and launching a proxy process, the skill introduces hidden browser-control infrastructure that users and reviewers would not expect, increasing the risk of unauthorized data access or future abuse.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

This script detects Chrome remote-debugging ports, prompts the user to enable remote debugging, and starts a local CDP proxy, which gives browser-inspection/control capability unrelated to a travel-duration skill. In this context, that is a dangerous overreach because it expands the skill's effective privileges into browser access and could expose tabs, cookies, page content, or authenticated sessions if reused by other components or misconfigured.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill performs shell execution, environment-dependent checks, and network/browser automation but does not declare any explicit tool scope such as permissions or allowed-tools. This creates an overbroad execution model where an agent may invoke powerful local capabilities without clear authorization boundaries, increasing the risk of unintended command execution or data access.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

L006 规定“只要涉及‘目的地 + 可用天数’的组合,就应该触发此 skill”,没有给出边界条件或排除场景。这会覆盖大量普通旅行聊天或随口比较的问题,缺少足够具体的触发约束,属于模糊触发描述。

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill depends on accessing content from a login-gated community platform but does not clearly warn the user that using their logged-in browser session may expose personal account data, personalized feeds, or other sensitive browsing information. This omission undermines informed consent and can lead users to authorize risky access without understanding the privacy consequences.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill instructs the agent to have the user enable Chrome remote debugging for the browser instance, which grants broad control over the user's active browser session beyond the travel-evaluation purpose. If followed, this could expose authenticated tabs, cookies, browsing data, and page contents from unrelated sites, making the requested capability disproportionate and dangerous.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instructions tell the user to enable remote debugging but provide no meaningful safety warning about the breadth of access this grants. Because remote debugging can expose full control over the browser instance, omitting a strong warning materially increases the chance of unsafe user enablement.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The skill transmits user-derived destination queries to an external website and does so via a local browser-control bridge, causing user input and browsing actions to be sent outside the agent environment. While destination keywords are not highly sensitive by default, this still constitutes external data transmission and, in context, may also leverage the user's authenticated browser state.

Content

Scanner excerpt · SKILL.md (reported line 50)May include surrounding context.

bash
# 启动搜索页(将 KEYWORD 替换为实际关键词)
curl -s "http://localhost:3456/new?url=https://www.xiaohongshu.com/search_result?keyword=KEYWORD&source=web_search_result_note"

搜索关键词策略 -- 用多组关键词获取更全面的信息:

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 72)May include surrounding context.

可以用 /scroll 向下滚动加载更多结果:

bash
curl -s "http://localhost:3456/scroll?target=TARGET_ID&direction=bottom"

第三步:逐篇提取笔记正文

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The inline comment explicitly states that the proxy controls the user's everyday Chrome, which directly conflicts with the skill's declared holiday-sufficiency function. This mismatch is a strong signal of deceptive packaging: the code advertises or hides powerful browser-control behavior under an unrelated benign skill category, increasing the likelihood of abuse and reducing user awareness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The clickAt and setFiles endpoints can perform real user-affecting actions in the browser, including triggering gesture-gated behaviors and setting local files on upload inputs, without any authentication, authorization, or confirmation flow. Combined with access to the user's existing browser session, this can be used to submit forms, upload sensitive files, or interact with trusted sites invisibly.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The screenshot endpoint can write captured browser content to an arbitrary filesystem path supplied by the request, with no validation or user warning. This creates a path for sensitive visual data exfiltration and unsafe local file writes, especially dangerous when coupled with the proxy's access to authenticated browser pages.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code contains natural-language comments and console output in Chinese, starting at L002 and repeated throughout the file, which effectively forces a specific language experience for users and maintainers. The file does not indicate that the skill is intended only for Chinese-speaking users or provide any opt-in or alternative locale.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

User-facing comments and operational messages consistently use Chinese, and the file does not indicate any language choice or opt-in for other locales. Under the policy, forcing a specific language without offering user choice can be a natural-language locale violation.

Content

No source excerpt is available for this finding.