Back to skill

Security audit

Visual Bug Hunter

Security checks for vulnerabilities and agentic risk

Overview

The skill appears intended for UI bug hunting, but it grants live UI automation and code-changing workflows without clear user confirmation boundaries.

Install only if you are comfortable with an agent using screenshots and live UI automation for a named app under test. Ask for confirmation before any clicking, typing, app launch, or code edit, and review diffs before applying them. If using the optional CLI from README.md, pin a reviewed version or commit and install it in an isolated environment.

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
README.md:39
Finding

Unpinned External Executable Dependencies

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 39 and 42–45
Vulnerability Type: Supply-chain exposure through unpinned external packages and mutable source repositories
Risk Level: Medium

Vulnerable Code Snippet

bash
pip install visual-bug-hunter
bash
git clone https://github.com/yitao2027/visual-bug-hunter.git
cd visual-bug-hunter
pip install -e .

Technical Analysis

The documented installation methods retrieve and execute Python packaging code from external sources without pinning an exact package version, immutable commit, cryptographic hash, or signature.

The PyPI package and GitHub implementation are not included in the audited artifact, which contains only README.md and SKILL.md. Consequently, the audit cannot verify package build hooks, transitive dependencies, runtime network behavior, filesystem access, screenshot processing, subprocess use, or other executable behavior.

The GitHub command clones the repository's default branch, whose contents can change after this Skill has been reviewed. The unversioned PyPI command likewise installs whichever release the package index currently resolves. During installation, Python build backends and package setup hooks may execute code with the installing user's privileges.

This is a supply-chain weakness rather than evidence that the currently referenced package is malicious.

Attack Path

  1. An attacker compromises the referenced PyPI project, a maintainer account, the GitHub repository, or a relevant dependency.
  2. The attacker publishes a modified package release or changes the repository's default branch.
  3. A user follows one of the documented installation procedures.
  4. pip downloads and processes attacker-controlled package metadata, build configuration, dependencies, or installation hooks.
  5. Malicious code executes with the user's privileges during installation or later when the installed CLI is invoked.
  6. The payload may access resources ava ...[truncated 771 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the PyPI dependency to a reviewed, exact version rather than installing the latest available release.
  2. Generate a locked dependency set and require cryptographic hashes, such as through pip --require-hashes.
  3. Pin source installations to a reviewed immutable commit instead of cloning a mutable default branch.
  4. Publish and verify signed releases, attestations, checksums, and build provenance.
  5. Include the executable implementation in the audited artifact so its installation hooks, dependencies, and runtime behavior can be reviewed.
  6. Review and pin all transitive dependencies, not only the top-level package.
  7. Install the tool in an isolated, least-privileged environment and grant screen capture or accessibility permissions only when required.
  8. Ensure the executable implementation technically enforces the project-directory and target-process restrictions declared in SKILL.md.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
78% confidence
Finding

The description says the tool can "click" programmatically, and the usage section later says it will automatically start an app and capture its state. For a markdown file, this is a system-integrity-relevant behavior that should be disclosed so users understand the skill may launch applications and perform automated UI actions on their machine.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The markdown states the tool will "Generate minimal diffs" and "Verify" fixes worked, which implies code or file modifications to the user's project. The README does not include any warning about file-changing behavior, review requirements, or potential impact on user code/data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The trigger conditions are broad enough to activate on generic bug-fixing or self-check requests, which can cause the skill to engage UI automation in situations where the user did not clearly consent to screenshot capture, live clicking, or text entry. Because this skill is designed to interact with running applications, overbroad activation increases the chance of unintended actions in the user's environment.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill describes automated clicking and text input but does not require an explicit pre-action warning or user confirmation before performing live UI actions. This is dangerous because automated interaction with a real desktop or web app can submit forms, modify data, trigger purchases, send messages, or alter system/application state unintentionally.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The file is predominantly in English but includes a Chinese-only promotional block at the top, with no opt-in or explanation for the language switch. This can violate a language/locale policy requiring user choice or consistent locale handling.

Content

No source excerpt is available for this finding.