Back to skill

Security audit

ChatGPT Ads Landing

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this skill suspicious because scanners report a consent bypass and incomplete inspection of several files, alongside other security and scope concerns.

Consider the consent bypass, public-access guidance, and preview-server weakness before installing. The scan was partial, so treat its findings as incomplete evidence rather than assurance that unreported behavior is safe.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
assets/engine/scripts/dev.js
Finding

Preview server permits directory traversal into sibling paths sharing the dist prefix

Content
View full analysis

Impact

A reachable attacker can retrieve existing files under sibling paths such as dist-backup or dist-private without authentication. The exposure is limited to paths sharing the checked prefix; no specific secret file was present or demonstrated.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (14)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
dynamic (country routing, pixel events, consent) is progressive enhancement in `lp.js`.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 74)May include surrounding context.

md
`publish.mjs ship` / `package.js` rewrite `page.json` (url, siteId) and rebuild at the same

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 77)May include surrounding context.

md
kill dev.js, run `node scripts/build.js` once, `check.js` must pass, then ship.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 103)May include surrounding context.

md
Run `verify.mjs` first; answer from its output. For a fuller content review add

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/platform-deploy.md (reported line 50)May include surrounding context.

md
| Step | Endpoint |
|---|---|
| login (once) | `POST /v1/open-api/device/code {app_id}` → user approves at `verification_url` → poll `POST /v1/open-api/device/token` (errCode `AUTHORIZATION_PENDING` until approved) → api-key saved to `~/.config/chatgpt-ads-landing/credentials.json` (0600) |
| upload | `GET /v1/site-hosting/site/deploy-credential` → STS key + one-off `dir` (`staging/{owner}/{nonce}/`) → PUT each file of `dist/<slug>/` to COS at `dir + relative path` |
| first deploy | `POST /v1/site-hosting/site {name, subdomain_hint, entry, password_enabled:false, pending_dir, version_remark}` → `site_id`, `site_url` |
| new version | `PUT /v1/site-hosting/site {site_id, name, entry, password_enabled, password, pending_dir, version_remark}` — always send name and password fields: omitted fields are treated as "clear" |

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/lib/platform.mjs (reported line 4)May include surrounding context.

js
/**
 * Hosting-platform client for the skill's optional publish script — zero deps, Node ≥ 22.
 *
 *   auth      device-code login → api-key, stored in ~/.config/chatgpt-ads-landing/credentials.json (0600)
 *   api       JSON calls with the `api-key` header; unwraps BaseResponse {errId, errCode, errMsg, data}
 *   cos       PUT one object to Tencent COS with the STS credential from site/deploy-credential
 *

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/lib/platform.mjs (reported line 22)May include surrounding context.

js
// No default: set BV_API_BASE (or --api-base) to the API address of your hosting platform.
export const DEFAULT_API_BASE = '';
const CRED_FILE = process.env.BV_CREDENTIALS || join(homedir(), '.config', 'chatgpt-ads-landing', 'credentials.json');

export const sleep = (ms) => new Promise((r) => setTimeout(r, ms));

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The activation list includes generic deployment and routing phrases without explicitly requiring an OpenAI/ChatGPT Ads context.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The instruction to activate whenever someone mentions “一键部署 / 自动上线” (one-click deployment / automatic publishing), “落地页体检” (landing-page health check), or “按国家/IP 跳转” (country/IP routing) extends beyond the stated advertising domain. These phrases also describe unrelated hosting and website tasks, so treating them as independent triggers could invoke this skill unintentionally.

Content

No source excerpt is available for this finding.

The no-JavaScript pixel bypasses the advertised UK/EU consent handling.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The build unconditionally inserts a measurement image whenever a pixel ID exists, including on pages configured to require consent. With JavaScript disabled, the consent controls cannot run, but the browser still transmits a page-view request and network metadata to OpenAI, bypassing the advertised consent protection.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · assets/engine/shared/lp.js (reported line 186)May include surrounding context.

js
// ---- IP country (routing.ipLookup) ----
  // A static site cannot see the visitor IP, so ask public lookup endpoints,
  // in order, each with a short timeout. Both allow any origin (CORS *).
  var IP_SOURCES = [
    ['https://www.cloudflare.com/cdn-cgi/trace', function (t) { var m = /^loc=([A-Z]{2})$/m.exec(t); return m ? m[1] : ''; }],
    ['https://api.country.is/', function (t) { try { return JSON.parse(t).country || ''; } catch (e) { return ''; } }]

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 · assets/engine/shared/lp.js (reported line 189)May include surrounding context.

js
// in order, each with a short timeout. Both allow any origin (CORS *).
  var IP_SOURCES = [
    ['https://www.cloudflare.com/cdn-cgi/trace', function (t) { var m = /^loc=([A-Z]{2})$/m.exec(t); return m ? m[1] : ''; }],
    ['https://api.country.is/', function (t) { try { return JSON.parse(t).country || ''; } catch (e) { return ''; } }]
  ];
  var IP_TIMEOUT_MS = 2000;

The instruction to turn off hosting password protection lacks a warning about exposing protected content publicly.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The file directs users to disable password protection so the advertising crawler can access a landing page, but does not explain that this also permits public access. Applying this instruction to a protected deployment could expose unpublished content or other material not intended for public release.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · scripts/verify.mjs (reported line 249)May include surrounding context.

js
async function groupConsent(cfg) {
  const countries = (cfg.consent && cfg.consent.countries) || [];
  console.log('\n■ consent — cookie banner flow');
  if (!countries.length) { report('consent', 'info', 'no consent.countries configured'); return; }
  const acceptText = cfg.consent.accept || 'Accept';
  const declineText = cfg.consent.decline || 'Decline';
  const pe = cfg.pixelEvents || {};