Back to skill
Skillv1.0.0

ClawScan security

Pipeworx art · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

ReviewApr 7, 2026, 5:14 PM
Verdict
Review
Confidence
high
Model
gpt-5-mini
Summary
The skill claims to talk to the Met's open API but relies on an external Pipeworx gateway and an npx-installed remote helper (not declared), which creates a risk of running unreviewed remote code and is inconsistent with the declared requirements.
Guidance
This pack appears to provide Met Museum data, but it routes calls through an external Pipeworx gateway and recommends using 'npx -y mcp-remote@latest' — which will download and execute code from npm at runtime. Before installing: (1) confirm you trust https://gateway.pipeworx.io and the pipeworx project; (2) verify the mcp-remote package source (review its code/release) and add Node/npx to declared requirements; (3) consider requiring the skill to call the Met's official API (collectionapi.metmuseum.org) directly if you prefer no third-party proxy; (4) run this skill in a sandboxed environment or restrict agent permissions if you allow it; (5) if you are unsure about executing remote npm packages, do not enable autonomous invocation or avoid installing the skill.

Review Dimensions

Purpose & Capability
concernThe description says it connects to the Met's open-access API, but the runtime example and setup point to https://gateway.pipeworx.io/art/mcp (a proxy/gateway) rather than the Met's official endpoints. The SKILL.md's Setup also instructs using npx to run mcp-remote, yet 'npx' / Node is not listed in the required binaries — a clear mismatch between claimed dependencies and actual instructions.
Instruction Scope
concernInstructions include POSTing JSON-RPC calls to a third-party gateway and recommend configuring an mcpServer that will run 'npx -y mcp-remote@latest https://gateway.pipeworx.io/art/mcp'. That effectively instructs the agent/runtime to fetch and execute code from npm and to use a remote service as the tool backend. The SKILL.md does not instruct reading local files or secrets, but it does delegate runtime behavior to an external, unreviewed component.
Install Mechanism
concernThere is no formal install spec, but the Setup suggests invoking npx to fetch and run mcp-remote@latest from npm. Using npx downloads and executes package code at runtime from the public registry — moderate-to-high risk if the package or the gateway is untrusted. The Skill fails to declare Node/npx as a required binary, which is inconsistent and could cause unnoticed remote code execution.
Credentials
noteThe skill declares no required environment variables or credentials, which is proportionate for a read-only art catalog. However, because it routes requests through a third-party gateway and a remote npm tool, additional implicit network access and potential token use could occur but are not declared.
Persistence & Privilege
concernalways is false (good), but autonomous model invocation is allowed (default). Combined with the setup that causes the agent/runtime to fetch and run mcp-remote from npm and to send JSON-RPC to an external gateway, this increases the blast radius: the agent could autonomously run unvetted remote code and communicate with a third-party service.