Back to skill

Security audit

IceCube Social Ops

Security checks for vulnerabilities and agentic risk

Overview

This skill is for social-media automation, but it asks for broad logged-in browser control and defaults to automatic public posting and message handling after one-time authorization.

Install only if you intentionally want an agent to operate real social accounts. Use a dedicated browser profile or isolated account, avoid normal browsing sessions, require manual approval for posts/replies/follows at first, do not save raw cookies unless you understand the risk, and avoid logging private messages unless users have consented and retention is defined.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:31
Finding

Overprivileged Automation Through an Authenticated Chrome Debugging Session

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 31–35
Vulnerability Type: Excessive access to an authenticated browser profile
Risk Level: Medium

Vulnerable Code:

markdown
2. Enable Chrome remote debugging:
   ```bash
   /Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222
   ```
3. IceCube connects to the logged-in Chrome through the browser skill

Related instructions at lines 48–49 also tell the user to keep Chrome open or save cookies and allow the browser skill to publish automatically.

Technical Analysis

The Skill directs the user to expose an authenticated Chrome instance through the Chrome DevTools Protocol on port 9222. A client connected to this endpoint can exercise extensive control over the exposed browser, including inspecting pages, navigating tabs, executing JavaScript in page contexts, and interacting with authenticated sessions available to that browser profile.

The instructions do not require a dedicated browser profile restricted to X/Twitter and Xiaohongshu. If the user launches a general-purpose profile, the resulting authorization boundary can include unrelated tabs, sessions, and browser-accessible data. This violates least-privilege principles because social-media publishing does not require control over every authenticated context in the user's normal browser profile.

The Skill also defaults to autonomous publishing and interaction. Although these actions align with its declared social-media purpose, combining autonomous behavior with broad browser-profile access increases the consequences of compromised, malformed, or misdirected browser automation.

No malicious exploitation logic, credential-exfiltration command, or external payload was found in the reviewed file. Exploitation would require another process or automation component to obtain access to the debugging endpoint.

Attack Path

  1. The user signs into ...[truncated 1524 chars]
Remediation
View remediation

Remediation Suggestions

  1. Create a dedicated Chrome profile used exclusively for the two intended social-media accounts. Do not use the user's normal browsing profile.
  2. Bind the debugging service explicitly to the loopback interface where supported, and use host firewall rules to deny access to port 9222 from other devices, containers, and untrusted local users.
  3. Prefer platform APIs with narrowly scoped OAuth permissions instead of browser-wide automation whenever practical.
  4. Run the browser under a dedicated, low-privilege operating-system account or isolated environment.
  5. Close unrelated tabs and ensure that no unrelated accounts are authenticated in the automation profile.
  6. Avoid exporting or persistently saving raw cookies. If session persistence is necessary, protect the profile with restrictive filesystem permissions and define a rotation and revocation process.
  7. Require explicit human approval for initial posts, private-message responses, account changes, follows, and other high-impact actions.
  8. Restrict the browser skill to allowlisted domains and block navigation to all other origins.
  9. Shut down the debugging-enabled browser immediately after the automation task and document how users can revoke sessions if compromise is suspected.
Vulnerability Patterns
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrases are broad and map to common requests about account management, posting, and social operations. This increases the chance the skill activates in situations where the user did not intend a high-impact automation workflow that can post externally or access authenticated sessions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill normalizes 'authorize once, then auto-operate' and defaults to automatic posting/interaction without prominent warnings or guardrails. Because it can issue outward communications from real accounts, unintended activation or misgenerated content could immediately affect public reputation, platform compliance, and account safety.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly instructs enabling Chrome remote debugging and attaching to a logged-in browser session. That grants broad control over the user's authenticated browser context, potentially exposing cookies, session data, private content, and the ability to perform arbitrary actions beyond social posting. In this skill, that capability is especially risky because the workflow is designed for persistent automated account operation after one-time authorization.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The instruction to keep Chrome open or preserve cookies for ongoing automated use encourages persistent authenticated sessions. Persistent session reuse raises the likelihood of unauthorized account access, session theft, and silent long-term control over the user's social accounts.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

Recording private direct-message inquiries and service requests in memory or logs creates a data-retention surface for potentially sensitive personal or business information. Without clear consent, storage limits, and access controls, this can expose confidential communications and create privacy/compliance issues.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The daily report template explicitly summarizes private-message consultations and service-demand data, institutionalizing retention and secondary use of private communications. This increases privacy risk by turning sensitive interactions into durable analytics artifacts that may be broadly accessible or insufficiently protected.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The workflow expands from posting into monitoring private messages, logging service requests, and conversion tracking, which introduces collection and processing of potentially sensitive user communications. This broadens the data-access scope from content publishing into CRM-style surveillance without clear minimization, consent, or retention controls.

Content

No source excerpt is available for this finding.