Back to skill

Security audit

Tandem Browser

Security checks for vulnerabilities and agentic risk

Overview

This browser-control skill is mostly transparent, but it asks users to run a local browser with broad debugging powers and Chromium sandboxing disabled.

Review this carefully before installing. Use it only with a Tandem profile and OS account you are comfortable exposing to agent-driven browsing, avoid authenticated personal sessions unless necessary, prefer isolated sessions, revoke trust grants when done, and run Tandem in a container or otherwise restricted environment if browsing untrusted sites.

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:34
Finding

Chromium Sandbox Explicitly Disabled

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 34–36
Vulnerability Type: Insecure browser configuration
Risk Level: Medium

Vulnerable Code

markdown
**Rules:**
- ✅ `--no-sandbox` — always required on Linux
- ❌ `--disable-gpu` — **DO NOT USE** — breaks GUI

Technical Analysis

The skill explicitly requires launching the Electron/Chromium-based Tandem Browser with --no-sandbox. Chromium's sandbox is a defense-in-depth boundary that restricts compromised renderer processes and reduces their ability to interact with host resources.

Disabling this protection does not by itself execute malicious code, but it materially increases the severity of a browser-engine or renderer compromise. This is particularly relevant because the skill is designed to navigate arbitrary websites and process potentially hostile web content. The document does not prescribe an equivalent compensating control such as execution inside a restricted container or virtual machine.

Attack Path

  1. Tandem Browser is launched according to the skill's instructions with --no-sandbox.
  2. The agent navigates to an attacker-controlled or compromised website.
  3. The website exploits a vulnerability in Chromium, Electron, or a renderer-accessible browser component.
  4. Because the Chromium sandbox is disabled, normal renderer isolation is absent or substantially weakened.
  5. Exploit code accesses resources available to the Tandem process under the current operating-system account.

Successful exploitation depends on a separate browser vulnerability; the documented flag weakens containment rather than independently providing code execution.

Impact Assessment

A successful browser compromise could expose files, credentials, browser state, authentication sessions, and other resources accessible to the operating-system account running Tandem. It could also permit arbitrary actions with that account's privileges.

The potential scope is the user's account and ...[truncated 181 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the unconditional requirement to launch Tandem with --no-sandbox.
  2. Configure the host so Chromium can use a supported sandbox mechanism, such as unprivileged user namespaces or the appropriately installed setuid sandbox.
  3. If sandbox disablement is unavoidable, run Tandem in a dedicated container or virtual machine with:
    • A non-root, dedicated operating-system account.
    • A read-only or narrowly scoped filesystem.
    • No access to unrelated home-directory files, SSH keys, cloud credentials, or host browser profiles.
    • Restricted network access appropriate to the browsing task.
    • Dropped Linux capabilities and no-new-privileges.
    • Resource and process limits.
  4. Use isolated Tandem sessions for untrusted browsing and avoid inheriting authenticated session state unless required.
  5. Document why sandbox disablement is necessary, the affected platforms, and all mandatory compensating controls.
  6. Keep Electron and Chromium patched to reduce exposure to known renderer and browser-engine vulnerabilities.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (1)

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The manifest describes a shared-browser interaction skill for browsing, snapshotting, clicking, typing, navigating, and coordinating with the user. The documented tool surface also includes tandem_network_* for HAR/network log inspection and tandem_devtools_send for a CDP debug bridge, which are substantially broader debugging/instrumentation capabilities not justified by that stated purpose.

Content

No source excerpt is available for this finding.