[BUG] --agent flag in CLI does not allow all internal tools by default; `tools: Task` doesn't work

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

claude --agent <agent-type> does not have all internal tools on by default. Note that <agent-type> in this case is not specified with tools nor disallowedTools, though the issue seems to persist with disallowedTools: [] (no tools: settings).

Note: This is true regardless of --resume (i.e. --resume <chat-name> --agent <agent-type>)

The session chat seems to only have these tools (note I disabled WebSearch, you'll probably have it in your own list if not disabled):

- Bash
- Glob
- Grep
- Read
- Edit
- Write
- NotebookEdit
- Skill
- TaskCreate
- TaskGet
- TaskList
- TaskUpdate
- ToolSearch
- ListMcpResourcesTool
- ReadMcpResourceTool
- WebFetch

Full tools list from: https://code.claude.com/docs/en/settings.md

- AskUserQuestion
- Bash
- TaskOutput
- Edit
- ExitPlanMode
- Glob
- Grep
- KillShell
- MCPSearch
- NotebookEdit
- Read
- Skill
- Task
- TaskCreate
- TaskGet
- TaskList
- TaskUpdate
- WebFetch
- WebSearch
- Write
- LSP

UPDATE: tools: Task, ... still does not allow --agent session to use Task!

What Should Happen?

~~It's not clear this is a bug, but~~ the language in the docs is a big ambiguous: https://code.claude.com/docs/en/sub-agents
Perhaps, there should be a reliable way of allowing all tools by default, something like:

tools: all
disallowedTools: Edit, Task

Error Messages/Logs

Steps to Reproduce

1. Create a .claude/agents/basic-agent.md:

```md
---
name: basic-agent
description: "Use this basic agent to test weird behavior?"
model: sonnet
---

You are an awesome and disciplined general-purpose AI agent!
```

1. Run claude --agent basic-agent.md
2. Prompt: can you 1. tell me which tools are available to you? 2. try spawning a subagent with 'Task' tool?

Claude will likely respond with something about TaskList/etc. and propose to use Bash.
___

Update: Does not work even when specifying tools: Task:

---
name: basic-agent
description: "Use this basic agent to test weird behavior?"
model: sonnet
tools: Task, Read, Bash
---
...

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.29

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

_No response_