skills/yamadashy/skills/claude-exec/Gen Agent Trust Hub

claude-exec

Pass

Audited by Gen Agent Trust Hub on Jul 7, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is purely instructional and does not contain any executable scripts or hidden code. It provides documentation for using a legitimate CLI tool.
  • [COMMAND_EXECUTION]: The skill describes how to invoke the claude CLI via the Bash tool. It promotes security best practices by suggesting flags like --no-session-persistence, --bare, and --max-budget-usd to maintain a controlled execution environment.
  • [PROMPT_INJECTION]: The skill documents an interface for delegating tasks to a sub-agent, which is a surface for indirect prompt injection. 1. Ingestion points: User-provided <task> string interpolated into bash commands. 2. Boundary markers: The skill recommends using the -- flag for explicit argument separation. 3. Capability inventory: The sub-agent's capabilities are managed through flags like --allowedTools and --disallowedTools. 4. Sanitization: No specific input sanitization is defined within the skill itself. This attack surface is inherent to the tool's function and is documented with mitigation strategies.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 7, 2026, 12:54 AM
Security Audit — agent-trust-hub — claude-exec