[BUG] Sub-agents use tools without permission

Environment

• Platform (select one):
• [X] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version: 1.0.63 (Claude Code)
• Operating System: macOS 15.5
• Terminal: iTerm2

Bug Description

Claude code sub-agents can use tools even when none are permitted.

Steps to Reproduce

1. Create a new sub-agent using Claude Code CLI
2. Uncheck all tool permissions
3. Save sub-agent
4. Execute the sub-agent

Example:

---
name: prd
description: Use this agent PROACTIVELY to transform unstructured information (meeting notes, emails, chat logs, dictation transcripts, brainstorming sessions) into a formal Project Requirements Document (PRD). Can be used when: consolidating scattered requirements, formalizing verbal discussions, creating structured documentation from chaos, preparing requirements for technical analysis. Triggers: "create PRD", "formalize requirements", "structure these notes", "turn this into requirements". Always request the unstructured input documents before beginning synthesis.
tools: 
color: yellow
---

...

Expected Behavior

The sub-agent is unable to use tools because none are permitted.

Actual Behavior

Several tools were used by the sub-agent.

⏺ prd(Create PRD from ticket)
  ⎿  Done (7 tool uses · 23.8k tokens · 1m 55.9s)

Additional Context

Project-level Claude Code settings:

{
  "permissions": {
    "allow": [
      "Bash(say:*)"
    ],
    "deny": []
  }
}