agentic-awesome-skills-installer

Warn

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides commands to download software and content from an unverified third-party repository ("github.com/sickn33/agentic-awesome-skills") and an npm package ("agentic-awesome-skills"). This includes installation methods using git clone, curl, and npx.
  • [REMOTE_CODE_EXECUTION]: The skill encourages the execution of code downloaded at runtime via npx. It also provides Python and JavaScript examples that use subprocess.run or child_process to execute these external tools on the local system.
  • [COMMAND_EXECUTION]: The instructions include shell commands for the agent to modify the file system, such as rm -rf to delete directories and chmod -R u+w to change file permissions. It also instructs the agent to modify sensitive tool configuration files like .cursor/settings.json and ~/.claude/plugins/config.json.
  • [PROMPT_INJECTION]: The skill facilitates the mass installation of 1,935+ unvetted instruction files ("playbooks") from an external source, creating a large surface for indirect prompt injection where the agent may follow malicious instructions embedded in the downloaded files.
  • Ingestion points: The ~/.agents/skills directory and tool-specific plugin paths used by Claude, Cursor, and other assistants.
  • Boundary markers: No delimiters or instructions to ignore embedded commands are implemented for the downloaded content.
  • Capability inventory: The skill environment allows for shell execution, file system modification, and network operations.
  • Sanitization: No mechanism is provided to validate or sanitize the content of the thousands of downloaded skill files before they are ingested by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 8, 2026, 08:33 PM
Security Audit — agent-trust-hub — agentic-awesome-skills-installer