ethereum-app-builder
Warn
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The skill executes
npx create-eth@latest, which downloads and runs code from the npm registry. The packagecreate-ethis not maintained by a verified trusted organization listed in the security policy. - REMOTE_CODE_EXECUTION (MEDIUM): The use of
npxto download and immediately execute a remote package allows for unvetted code execution on the host system. - EXTERNAL_DOWNLOADS (MEDIUM): The skill is instructed to fetch and use reference skills from an untrusted GitHub repository (
technophile-04/ethereum-app-skill). This introduces a supply chain vulnerability where the agent's behavior can be altered by updates to the remote repository. - PROMPT_INJECTION (LOW): The skill exhibits an indirect prompt injection surface by treating the
AGENTS.mdfile (produced by the scaffolding tool) and third-party remote skills as 'sources of truth'. - Ingestion points:
<project-name>/AGENTS.mdandtechnophile-04/ethereum-app-skill/main/skills/ponder/SKILL.md. - Boundary markers: None present; the agent is told to follow the instructions 'strictly' and treat them as the 'source of truth'.
- Capability inventory: Subprocess execution via
npxand potentially other commands defined in the fetched skills. - Sanitization: None observed.
Audit Metadata