web-game-development
Fail
Audited by Gen Agent Trust Hub on Jul 28, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the script scripts/fetch-upstream-skills.sh to download content from https://github.com/MengTo/Skills.git. This repository is not a recognized trusted organization or well-known service, making the retrieval of code from this location a security risk.
- [REMOTE_CODE_EXECUTION]: The skill fetches and installs external SKILL.md files that define agent behavior. This effectively executes remote instructions. The script copies these files into the agent's local skills root, establishing a persistence mechanism where remote code is stored and reused across sessions.
- [COMMAND_EXECUTION]: The skill instructs the agent to run a bash script that performs directory creation, git operations, and recursive file copies to install external components.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface. 1. Ingestion points: Fetches and installs instruction sets from github.com/MengTo/Skills.git. 2. Boundary markers: Lacks integrity checks or delimiters for the downloaded content; it explicitly treats external files as the authoritative source of truth. 3. Capability inventory: Access to Bash, Read, Write, Edit, Glob, and Grep tools. 4. Sanitization: No validation or sanitization of the downloaded instruction files is performed before they are made available for agent execution.
Recommendations
- AI detected serious security threats
Audit Metadata