[BUG] Startup workspace-trust dialog does not warn about hooks in .claude/settings.json, while the /cd dialog does

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

The two trust dialogs disclose different things for the same directory.

Starting `claude` inside an untrusted directory whose `.claude/settings.json` defines a
`PreToolUse` command hook shows (2.1.270, verbatim):

```
Accessing workspace:
/Users/flo/claude-poc-sandbox/parent/sub
Quick safety check: Is this a project you created or one you trust? (Like your
own code, a well-known open source project, or work from your team). If not,
take a moment to review what's in this folder first.
Claude Code'll be able to read, edit, and execute files here.
Security guide
❯ No, exit
Yes, I trust this folder
```

No mention of hooks. Using `/cd` into the very same directory from a trusted session shows:

```
Moving to a new directory:
/Users/flo/claude-poc-sandbox/parent/sub
This session hasn't worked here before. Is this a directory you created or one you trust?
Claude Code'll be able to read, edit, and execute files here.
⚠ This directory configures hooks that run commands, declared in .claude/settings.json
These will apply to this session as soon as you move. Only proceed if you trust this configuration.
❯ No, stay put
Yes, move here
```

The startup dialog already inspects the settings file for pre-approved permissions (see #87012,
"This folder pre-approves 35 tool permissions in .claude/settings.local.json"), so the hook check
that `/cd` performs should be cheap to add there. Hooks are the more consequential of the two:
they run commands on the first tool call regardless of what the user later approves.

### What Should Happen?

The startup trust dialog should carry the same hook warning as the `/cd` dialog, naming the file
and ideally the event types and commands, e.g.:

```
⚠ This folder configures hooks that run commands, declared in .claude/settings.json
PreToolUse: /Users/flo/claude-poc-sandbox/fire.sh SUB_HOOK
These will run as soon as you start using tools. Only proceed if you trust this configuration.
```

### Error Messages/Logs

```shell
n/a (dialog text above)
```

### Steps to Reproduce

1. Create an untrusted repo with a command hook:
```sh
mkdir -p /tmp/hookrepo/.claude && cd /tmp/hookrepo && git init -q
cat > .claude/settings.json <<'JSON'
{"hooks":{"PreToolUse":[{"matcher":"","hooks":[{"type":"command","command":"touch /tmp/hookrepo/HOOK_FIRED"}]}]}}
JSON
```
Make sure `~/.claude.json` has no `projects["/tmp/hookrepo"]` entry.
2. `cd /tmp/hookrepo && claude` → dialog shown without any hook warning. Choose "No, exit".
3. `cd /tmp && claude`, trust `/tmp` if asked, then `/cd /tmp/hookrepo` → dialog shown **with**
the hook warning.
4. Accept in step 3, ask Claude to run `ls`, observe `/tmp/hookrepo/HOOK_FIRED` exists.
(Same would happen after accepting in step 2, without having been told.)

### Claude Model

Other

### Is this a regression?

I don't know

### Last Working Version

n/a

### Claude Code Version

2.1.270

### Platform

Anthropic API

### Operating System

macOS

### Terminal/Shell

iTerm2

### Additional Information

- Related: #87012 (same dialog, truncated permissions list), #92911 (trust prompt suppressed,
hooks silently disabled), #67319 (VS Code never shows the dialog).
- The `/cd` dialog's behavior is good and not documented; the docs for `/cd` (v2.1.246 notes)
say settings are re-read but not that trust is re-evaluated. Worth a line in the docs.

Start with the startup workspace-trust dialog and compare its handling of .claude/settings.json with the existing /cd dialog behavior. Reproduce the untrusted /tmp/hookrepo case from the issue, then trace the hook-warning check and its dialog text. Done means startup trust disclosure matches the /cd warning, including the configured hook information, with coverage for the reported flow.

Written by the indexing model from the issue text.