walrus-sites
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs users to fetch software and configuration from official and well-known sources as part of the standard setup process.
- Clones the portal repository from the official Mysten Labs GitHub organization (
github.com/MystenLabs/walrus-sites). - Pulls official Docker images for the portal server (
mysten/walrus-sites-server-portal). - Recommends using the official
suiuptoolchain manager to installsite-builderandwalrusCLI tools. - Fetches configuration from official Walrus documentation (
docs.wal.app). - [COMMAND_EXECUTION]: The instructions include standard command-line operations necessary for decentralized site deployment and local development.
- Provides commands for building projects (
npm run build), deploying content (site-builder deploy), and managing site lifecycles. - Includes utility commands for managing local server environments, such as identifying and terminating processes on specific network ports (
lsof,kill). - Outlines environment variable configuration for Docker containers to connect to Sui testnet and mainnet RPC nodes.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a system that processes external data, establishing an inherent attack surface for indirect instructions, though it is used here for the intended primary purpose.
- Ingestion points: The
site-buildertool reads content from a local directory (e.g.,dist/) provided by the user to upload to the Walrus network. - Boundary markers: The skill instructs the agent and user to use
ws-resources.jsonto define routing and headers, which helps maintain structure for processed content. - Capability inventory: The skill utilizes CLI tools with capabilities to read files, write configuration files, and perform network requests to blockchain fullnodes and storage aggregators.
- Sanitization: The workflow is a standard developer deployment process where the content source is controlled by the user, and the skill provides warnings against including sensitive data in the public storage network.
Audit Metadata