Back to skill

Security audit

CreBee | 多账号矩阵运营数据分析

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for CreBee account analytics, but its local client discovery flow can cause the agent to execute a locally supplied runtime without enough verification.

Review this skill before installing. It appears intended for legitimate CreBee analytics, but install it only if you trust the CreBee desktop client and your local machine environment; avoid using it on systems where another local process could impersonate the CreBee localhost service.

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
SKILL.md:61
Finding

Unauthenticated Local Runtime Discovery Can Redirect CLI Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 61–76
Vulnerability Type: Untrusted localhost discovery data used to select an executable
Risk Level: Medium

Relevant instruction excerpt:

sh
curl -s -m 2 --noproxy "*" http://127.0.0.1:3456/galic/v1/skill/cli-runtime

# Paths are derived from the executable location returned by discovery:
resources/crebee-cli/crebee.mjs
resources/ffmpeg/ffmpeg

# POSIX invocation form:
ELECTRON_RUN_AS_NODE=1 "<exe>" "<crebee.mjs>" <command-and-arguments>

Technical Analysis

The Skill instructs the Agent to query an unauthenticated HTTP endpoint on 127.0.0.1:3456 and use its response to identify the running CreBee executable. It then derives the packaged CLI path from that executable location and runs the CLI with ELECTRON_RUN_AS_NODE=1.

Although loopback limits remote network access, it does not authenticate the process listening on the port. The documented flow does not require verification of:

  • The identity of the process serving port 3456.
  • The authenticity or integrity of the discovery response.
  • The ownership and permissions of the returned executable.
  • Whether the resolved executable and CLI script remain inside a trusted installation directory.
  • A platform signature or cryptographic signature for the executable and CLI script.

Consequently, data received across an unauthenticated local network boundary controls which local executable and JavaScript entry point the Agent launches. Running Electron with ELECTRON_RUN_AS_NODE=1 gives the selected JavaScript code Node.js capabilities under the Agent's operating-system account.

This is an exploitable design flaw, but the reviewed files do not demonstrate that the Skill author intended to exploit it. The project is therefore classified as suspicious rather than malicious.

Attack Path

  1. The legitimate CreBee client is not running, leaving local port 3456 a ...[truncated 1604 chars]
Remediation
View remediation

Remediation Suggestions

  1. Authenticate the localhost discovery protocol using a per-installation or per-session secret that is not exposed through the discovery response.
  2. Cryptographically bind the returned runtime path to the installed CreBee client, such as by signing discovery responses and validating them against a trusted application key.
  3. Canonicalize all returned paths and require both the executable and CLI script to reside beneath an approved installation root.
  4. Verify file ownership and permissions before execution. Reject executables or scripts writable by untrusted users.
  5. Validate the platform code signature or a pinned cryptographic digest for both the CreBee executable and packaged CLI entry point.
  6. Confirm that the process listening on port 3456 is the expected signed CreBee process before trusting its response.
  7. Prefer an authenticated operating-system IPC mechanism over unauthenticated HTTP for local runtime discovery.
  8. Fail closed when identity or path validation fails; do not fall back to executing a path supplied solely by the localhost response.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The primary skill description is written as a Chinese-only invocation/use description ('想查看、分析自己账号的运营数据和作品表现时使用') with no indication that users may interact in other languages or that the skill is limited to Chinese-speaking contexts. Under the policy rule, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The entire troubleshooting guide is written as mandatory operational guidance in Chinese, with no indication that the user may choose another language or locale. Under the policy, forcing a specific language without opt-in is a natural-language policy violation unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.