--agent custom agent's disallowedTools policy is ignored in the interactive TUI (works in print mode)

What version of Kimi Code is running?

0.39.1

Which open platform/subscription were you using?

kimi code

Which model were you using?

k3

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

A custom agent file's `disallowedTools` policy is applied in print mode but silently ignored in the interactive TUI.

 With `~/.kimi-code/agents/dev.md` declaring `disallowedTools: [Read, Write, Edit]`, `kimi --agent dev` starts a TUI session without error, but the main

agent's tool list still contains Read/Write/Edit and the agent successfully executes Read on the first prompt.

 Verified via the session wire log (`agents/main/wire.jsonl` → `llm.tools_snapshot`):
 - TUI session started with `--agent dev`: 63 tools, Read/Write/Edit present
 - Print session started with `kimi --agent dev -p`: 60 tools, Read/Write/Edit absent
 - Print session without `--agent`: 63 tools (control)

What steps can reproduce the bug?

1. Create ~/.kimi-code/agents/dev.md:

 ---
 name: dev
 description: Default agent with Read/Write/Edit disabled
 disallowedTools:
   - Read
   - Write
   - Edit
 ---

 ${base_prompt}

 2. Run `kimi --agent dev` (interactive TUI) and prompt: read @AGENTS.md
    → the agent calls the Read tool and succeeds (unexpected).
 3. Compare with `kimi --agent dev -p "Reply with exactly: ok"`, then inspect the
    new session's wire.jsonl: Read/Write/Edit are absent from llm.tools_snapshot.

What is the expected behavior?

Per https://www.kimi.com/code/docs/en/kimi-code-cli/customization/agents.html, --agent works in both print mode and the interactive TUI, and an agent's
tools/disallowedTools "shape the tools shown to the model and are enforced again before execution". The TUI startup session should bind the qwen profile
and exclude Read/Write/Edit from the main agent's tool set, exactly as print mode does. An unknown agent name should fail at startup on both paths.

Additional information

No response

Contribution

• I am willing to submit a PR for this bug fix myself (please wait for maintainer approval in this issue first)

Start by comparing the startup paths for kimi --agent dev and kimi --agent dev -p, using the custom agent definition in ~/.kimi-code/agents/dev.md and each session's agents/main/wire.jsonl tools snapshot. Done means the interactive TUI excludes Read, Write, and Edit just like print mode, and an unknown agent name fails at startup in both modes.