add-backend-tool

Warn

Audited by Gen Agent Trust Hub on Jun 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions explicitly reference a run_terminal_command helper and suggest implementing tools for operations such as 'git operations' and 'listing directory contents'. This encourages the use of shell-executing functions which, if not strictly validated, can lead to arbitrary system command execution.
  • [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface where untrusted user input directly influences the generation of executable code.
  • Ingestion points: User prompts containing keywords like "new tool" or "add tool" trigger the skill's logic.
  • Boundary markers: The skill lacks specific instructions or delimiters to isolate user-provided tool requirements from the agent's code generation logic.
  • Capability inventory: The skill leverages file modification (writing to backend/main.py) and encourages the use of subprocess execution via run_terminal_command.
  • Sanitization: The instructions provide only a general advisory to "Consider security implications" without requiring specific input validation or sanitization logic for the new tools being created.
  • [DYNAMIC_EXECUTION]: The primary purpose of the skill is to have the agent perform self-modification of its source code (backend/main.py). It provides templates for generating Python functions and WebSocket handlers that are dynamically incorporated into the system's execution flow.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 20, 2026, 04:30 AM
Security Audit — agent-trust-hub — add-backend-tool