claude-code-course
Warn
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION] (MEDIUM): The skill performs shell operations to manage progress and demonstrate features. In 'skills/lessons/lesson-4.md', the tutor instructions command the AI to execute 'rm -rf .claude/settings.json' as part of a cleanup process. This is a destructive operation on a standard configuration file and may lead to unintentional data loss.
- [EXTERNAL_DOWNLOADS] (LOW): The skill references and recommends several external packages (e.g., 'fastmcp', '@anthropic/mcp-filesystem') and provides a git clone URL for the course repository. These are standard development resources but are not all from the specifically listed trusted organizations.
- [PROMPT_INJECTION] (LOW): The 'tutor' agent defined in 'agents/tutor.md' uses the 'Read' and 'Glob' tools to inspect the codebase and answer questions. It lacks specific delimiters or instructions to ignore embedded commands in the files it reads, leaving it susceptible to indirect prompt injection from data in the workspace.
Audit Metadata