Refua

PassAudited by ClawScan on May 1, 2026.

Overview

Refua is a coherent, purpose-aligned instruction-only skill, with the main user consideration being that it asks you to install and run external Python/MCP components and model assets.

Before installing, verify the refua and refua-mcp packages and consider using a virtual environment or isolated machine, especially if working with confidential molecular designs. The artifact does not show malicious behavior, credential use, or hidden persistence.

Findings (3)

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.

What this means

Installing the skill’s dependencies may bring third-party code and model files onto the user’s machine.

Why it was flagged

The skill asks the operator to install external, unpinned Python packages and download model/assets. This is central to the stated purpose, but it creates normal package/provenance risk that users should verify.

Skill content
`pip install refua[cuda]` ... `pip install refua` ... `pip install refua-mcp` ... `python -c "from refua import download_assets; download_assets()"`
Recommendation

Install only from trusted package sources, consider pinning versions, review the referenced project, and use an isolated environment if handling sensitive research data.

What this means

Running the server executes local Python package code and may consume CPU/GPU resources while active.

Why it was flagged

The skill requires running an installed local MCP server process. This is expected for the MCP-based workflow and is user-directed, not hidden or automatic.

Skill content
Start the server using the module entrypoint: `python3 -m refua_mcp.server`
Recommendation

Run the server only after verifying the installed package, and stop it when the workflow is complete.

What this means

Potentially proprietary molecular structures or sequences may be provided to the MCP server during normal use.

Why it was flagged

The agent will pass protein, ligand, SMILES, or related task inputs through an MCP tool boundary. This is purpose-aligned, but users should understand where confidential discovery inputs are being processed.

Skill content
This skill runs and connects to the **refua-mcp** MCP server, which exposes Refua’s “unified Complex API” as MCP tools
Recommendation

Use a trusted local server configuration and avoid sending confidential research inputs unless you understand the server’s storage, logging, and network behavior.