Back to skill

Security audit

Xiaopi Agent Browser

Security checks for vulnerabilities and agentic risk

Overview

This browser automation skill is mostly coherent, but it documents reusable login-session storage and sensitive browser capture features without enough safety guidance.

Review this carefully before installing. Use it only for sites and accounts you intend to automate, avoid saving authenticated state unless necessary, treat state files and recordings as secrets, clean them up after use, and prefer a pinned or isolated install instead of a global latest-version install.

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:16
Finding

Unpinned Third-Party Package and Source Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 16-26; CONTRIBUTING.md, lines 15-18
Vulnerability Type: Unpinned and unverifiable third-party dependency installation
Risk Level: Medium

Vulnerable Code

SKILL.md, lines 16-18:

bash
npm install -g agent-browser
agent-browser install
agent-browser install --with-deps

SKILL.md, lines 23-26:

bash
git clone https://github.com/vercel-labs/agent-browser
cd agent-browser
pnpm install
pnpm build

CONTRIBUTING.md, lines 15-18:

bash
1. Install the latest version
   ```bash
   npm install -g agent-browser@latest
text

### Technical Analysis

The installation instructions retrieve and execute third-party package content without pinning an audited package version, Git commit, or integrity digest. The explicit use of `@latest` further directs users to install whichever release is current at execution time.

NPM installation and source builds can execute lifecycle or build scripts supplied by the downloaded package and its transitive dependencies. A global installation may also place executable files into shared user-level or system-level binary locations, depending on the NPM configuration and privileges used.

The Git-based installation similarly follows the repository's default branch and resolves dependencies through `pnpm install` without identifying a reviewed commit or documented lockfile verification procedure. Consequently, the effective code installed by users can change after this skill package has been audited.

This finding does not establish that the current upstream package is malicious. The vulnerability is the absence of reproducible version pinning and integrity controls, which creates a supply-chain attack opportunity.

### Attack Path

1. An attacker compromises the upstream NPM package, its maintainer account, the source repository, or a transitive dependency.
2. The attacker publishes a malicious release, modifies the default branch, or introduces
...[truncated 1112 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin agent-browser to a specifically reviewed version rather than using an implicit current release or @latest:

    bash
    npm install --global agent-browser@<reviewed-version>
    
  2. Publish and verify the expected NPM integrity digest or signed provenance for the approved package artifact.

  3. For source installation, pin the repository to a full reviewed commit hash:

    bash
    git clone https://github.com/vercel-labs/agent-browser
    cd agent-browser
    git checkout <reviewed-full-commit-hash>
    
  4. Require a committed lockfile and use a frozen or immutable installation mode so dependency resolution cannot silently change:

    bash
    pnpm install --frozen-lockfile
    
  5. Avoid global installation where possible. Prefer a project-local installation or an isolated container with restricted filesystem and network permissions.

  6. Explicitly warn users not to run installation commands with sudo or administrative privileges.

  7. Add an update process that reviews new versions, dependency changes, lifecycle scripts, and build scripts before changing the pinned version.

  8. Ensure SKILL.md and CONTRIBUTING.md reference the same reviewed version and integrity verification procedure.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · CONTRIBUTING.md (reported line 60)May include surrounding context.

md
## Adding New Commands to the Skill

Update SKILL.md when the upstream CLI adds new commands.
- Keep the Installation section
- Add new commands in the correct category
- Include usage examples

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · SKILL.md (reported line 41)May include surrounding context.

agent-browser open # Navigate to page agent-browser snapshot -i # Get interactive elements with refs agent-browser click @e1 # Click element by ref agent-browser fill @e2 "text" # Fill input by ref agent-browser close # Close browser

text

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The documented screenshot, PDF, video recording, upload, and state-saving features can write local files or capture sensitive on-screen/account data, yet the skill gives no caution about filesystem modification or secret capture. In an autonomous agent workflow, this increases the risk of collecting or persisting credentials, personal data, internal pages, or proprietary content without user awareness.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly documents setting HTTP basic-auth credentials and saving/loading browser session state, but provides no warning that these values may contain passwords, cookies, or bearer-equivalent session material. In an agent context, this can lead to persistent storage, reuse, or accidental disclosure of secrets that grant account access without re-authentication.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
93% confidence
Finding

The skill exposes session persistence through saving and loading browser state, which commonly includes authentication cookies and other session tokens. Without guardrails or warnings, an agent may persist reusable authenticated state to disk, enabling account takeover or lateral access if the file is exposed or reused inappropriately.

Content

Scanner excerpt · SKILL.md (reported line 236)May include surrounding context.

bash
agent-browser state save auth.json    # Save session state
agent-browser state load auth.json    # Load saved state

Example: Form submission

Session Persistence

Medium
Category
Rogue Agent
Confidence
95% confidence
Finding

The authentication example normalizes saving login state to 'auth.json' and reloading it later, effectively encouraging storage of reusable authenticated sessions. In the context of an agent skill, examples strongly influence behavior, so this increases the likelihood of unsafe token persistence and replay across runs.

Content

Scanner excerpt · SKILL.md (reported line 265)May include surrounding context.

agent-browser wait --url "/dashboard" agent-browser state save auth.json

Later sessions: load saved state

agent-browser state load auth.json agent-browser open https://app.example.com/dashboard

text