Brex

PassAudited by VirusTotal on May 11, 2026.

Overview

Type: OpenClaw Skill Name: brex Version: 1.0.5 The skill provides a legitimate integration for the Brex financial platform using the Membrane CLI. It contains standard instructions for the agent to install the '@membranehq/cli' package, authenticate via OAuth, and manage corporate expenses and cards through the Membrane middleware. The logic is transparent, well-documented, and aligns with the stated purpose of spend management without any indicators of malicious intent or data exfiltration.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

ConcernMedium Confidence
ASI02: Tool Misuse and Exploitation
What this means

If the agent chooses the wrong action or parameters, it could alter Brex business records or initiate financial workflows such as transfers.

Why it was flagged

The skill exposes a generic action runner for Brex actions, including destructive and financial-transfer actions, but the provided visible instructions do not require explicit user confirmation or bounded execution for those high-impact operations.

Skill content
| Delete Vendor | delete-vendor | Deletes a vendor by ID. | ... | Create Transfer | create-transfer | Creates a new transfer. | ... membrane action run <actionId> --connectionId=CONNECTION_ID --json
Recommendation

Require explicit user confirmation for every create, update, delete, card, vendor, budget, expense, or transfer action; prefer read-only use unless the user clearly authorizes a specific mutation.

ConcernMedium Confidence
ASI03: Identity and Privilege Abuse
What this means

Installing and using the skill may give the agent persistent access to sensitive Brex account data and actions under the connected user's permissions.

Why it was flagged

The skill relies on delegated Brex access through Membrane with automatic credential refresh, but the provided artifact does not show least-privilege scopes, token lifetime, or revocation guidance.

Skill content
Membrane handles authentication and credentials refresh automatically ... membrane connection ensure "https://brex.com" --json
Recommendation

Use a least-privilege Brex/Membrane connection, review granted scopes carefully, and revoke the connection when it is no longer needed.

What this means

Your Brex data and authorization flow depend on Membrane's service boundary and security practices.

Why it was flagged

Membrane is a disclosed intermediary for Brex authentication and actions, so sensitive financial data and credentials may pass through or be managed by that provider.

Skill content
This skill uses the Membrane CLI to interact with Brex. Membrane handles authentication and credentials refresh automatically
Recommendation

Use this only if you trust Membrane for Brex access, and review Membrane's data handling and account security settings.

What this means

Future CLI versions could behave differently from the version reviewed here.

Why it was flagged

The setup uses a global npm install with the moving @latest tag. This is central to the skill's purpose, but it means the installed code can change over time and is not pinned by the skill artifact.

Skill content
npm install -g @membranehq/cli@latest
Recommendation

Install from a trusted source and consider pinning a reviewed Membrane CLI version.

NoteHigh Confidence
ASI01: Agent Goal Hijack
What this means

A remote response could influence how the agent proceeds during connection setup.

Why it was flagged

The skill tells the agent that remote connection responses may include agent-facing instructions. That can be useful for setup, but such instructions should not override the user's request or higher-priority safety rules.

Skill content
`clientAction.agentInstructions` (optional) — instructions for the AI agent on how to proceed programmatically.
Recommendation

Treat provider-returned instructions as untrusted task data and confirm any sensitive or mutating action with the user.