Fabric
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Command Execution (MEDIUM): The skill workflows execute local shell commands such as 'cat' and 'ls', and call an external CLI tool ('fabric') to handle YouTube transcripts and URL fetching.
- Dynamic Code Generation (MEDIUM): Patterns like 'create_coding_feature' and 'analyze_email_headers' are designed to generate functional bash scripts and file modifications based on untrusted input data. If an agent executes these outputs without human intervention, it could lead to arbitrary command execution.
- File System Modification (MEDIUM): The 'create_coding_feature' pattern implements a structured protocol for modifying and creating local files, which could be abused via indirect prompt injection if the agent automates the change application.
- External Downloads (MEDIUM): The skill requires the installation of external Go binaries ('code_helper') and references external system prompts via 'curl' from GitHub repositories.
- Indirect Prompt Injection (LOW): The skill is heavily optimized to process external content (articles, YouTube transcripts, security reports), making it vulnerable to embedded instructions that might influence its analytical output or command generation.
- Network Side-channel (LOW): Every workflow execution triggers an automated POST request to 'localhost:8888' to send notifications, which transmits metadata about skill usage.
Audit Metadata