Back to skill

Security audit

tencent cvp skill

Security checks for vulnerabilities and agentic risk

Overview

This skill openly controls an Android device through ADB, but its app-install guidance allows broad web and APK-search fallbacks without safety checks.

Install only if you intend to give the agent broad ADB control over a Tencent Cloud Virtual Phone. Treat app installation as sensitive: use trusted app stores or verified first-party sources, avoid arbitrary APK search results, and require explicit confirmation before installing or entering sensitive account information.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:177
Finding

Untrusted APK Acquisition Through Arbitrary Web Search

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 177-182
Vulnerability Type: Insecure third-party application acquisition
Risk Level: Medium

markdown
### When MyApp Fails

- Try browser: `adb shell am start -a android.intent.action.VIEW -d "https://official-site.com"`
- Last resort: web search for APK download

Technical Analysis

The skill recommends using a web search to locate an APK when installation through MyApp fails. Search results are not a trusted software distribution channel and may direct the Agent to counterfeit websites, repackaged applications, or malicious APK files.

The instructions do not require validation of the publisher domain, Android package name, signing-certificate fingerprint, cryptographic checksum, or repository provenance before installation. The example domain is also a generic placeholder rather than a verified vendor endpoint. Consequently, an attacker who controls or promotes a search result could supply an APK whose displayed application identity resembles the intended software while its executable content has been modified.

This is a supply-chain weakness because the effective application payload may originate from an arbitrary and unverified third party.

Attack Path

  1. Installation through MyApp fails or is unavailable.
  2. The Agent follows the documented fallback and searches the web for an APK.
  3. An attacker-controlled or compromised download page appears in the results.
  4. The page supplies a counterfeit or modified APK using the expected application name and branding.
  5. Without mandatory signature, checksum, and provenance checks, the APK may be accepted for installation.
  6. The malicious application executes within the Android environment and requests or abuses any permissions granted to it.

Impact Assessment

Successful exploitation can result in arbitrary malicious application code executing with the Android application sandbox privileges and any additional permissions ...[truncated 600 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the instruction to search arbitrary websites for APK downloads.
  2. Restrict downloads to the application's verified publisher domain or an explicitly approved repository using HTTPS.
  3. Maintain an allowlist containing the expected package name and publisher signing-certificate SHA-256 fingerprint.
  4. Verify the APK signing certificate before installation and reject mismatches or unsigned packages.
  5. Where reproducible release hashes are available, compare the APK against a pinned SHA-256 digest obtained through a separate trusted channel.
  6. Require explicit user approval before sideloading any application outside the designated app store.
  7. Record the source URL, package identity, version, checksum, and signing identity for auditability.
  8. Fail safely when provenance cannot be verified rather than continuing with installation.
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The description is very broad ('all-in-one' automation for any task involving operating the Android device), which can cause the skill to activate for a wide range of routine requests without clear boundaries. In a high-privilege ADB skill, overbroad activation is risky because it may be selected for actions involving app control, installation, navigation, or data exposure beyond what is necessary.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly instructs launching arbitrary external URLs from the device, which expands its capability from local ADB automation into internet navigation and content retrieval. In the context of a device-control skill, this creates a path for opening untrusted sites, initiating downloads, or driving the user into broader web workflows not constrained by the declared scope.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill documents app installation workflows, including fallback to browser and APK/web search, but provides no safety guardrails around source trust, sideloading, permissions, or device impact. Because this skill can directly automate installation steps on an Android device, missing warnings and restrictions substantially increase the chance of unsafe or unauthorized software being installed.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill advises searching the web for package names and, as a last resort, searching for APK downloads, which materially adds software acquisition from potentially untrusted sources. For an Android automation skill with install capability, this is dangerous because it can facilitate sideloading malicious or tampered apps onto the device.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The instructions explicitly direct handling for Chinese text input and use Chinese example content, but do not offer any user language or locale choice. This can constitute a language/locale policy issue if the skill assumes a specific language workflow without opt-in.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.