awesome-claude-code-subagents
Fail
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides standalone installation instructions (Option 4) that use
curlto download a shell script (install-agents.sh) and execute it directly. This bypasses security review and allows for arbitrary code to run on the user's system.\n- [COMMAND_EXECUTION]: The installation process involves manual execution of downloaded scripts (./install-agents.sh) and modifying file permissions (chmod +x), which is a high-risk activity when the source is not verified.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves its primary components and subagents from theVoltAgentGitHub repository. These external resources are downloaded at runtime or during installation without integrity verification or version pinning.\n- [PROMPT_INJECTION]: The installation of 131+ subagents introduces a significant surface for indirect prompt injection. These subagents are markdown files containing instructions that the AI agent follows. Malicious content in these external files could override safety guidelines or perform unintended actions.\n - Ingestion points: Remote markdown files (subagents) downloaded from
githubusercontent.comto~/.claude/agents/(SKILL.md).\n - Boundary markers: Absent. No delimiters or warnings are used to isolate the subagent instructions from the primary agent logic.\n
- Capability inventory: The skill facilitates the use of these agents within the Claude Code CLI, which has permissions for shell command execution and file system access (SKILL.md).\n
- Sanitization: None. The skill does not validate or sanitize the content of the downloaded subagents before they are utilized by the AI.
Recommendations
- AI detected serious security threats
Audit Metadata