Admin Debug Context

<!-- rules:stamp -->

Rules from common-docs/systems/platform/ui-shell/builder-panel/FEATURE.md — edit there, never here (python3 common-docs/meta/scripts/one_home.py --write).

1. No AI integration starts without a Builder feed. The same change that adds a data step to a surface publishes its inputs, findings, workflow step outputs, AI runs and provision snapshots, so Arman can see every value the page has, in real time, from the first minute. Tell him exactly what to click: one click to open the panel is fine, not knowing it exists is not.
2. The panel shows the truth, unmodified. Every exact request and response (client calls and the server's external calls), an Activity timeline, and derived values only in a separate group that links to its raw source. Values render through the canonical multi-format JSON component; copy is icon-only, never a "Copy" label.
3. Evidence is tiered by certainty and risk. Early raw-input steps are safe, read-only searches; deeper or risky steps fire only after a person confirms; a node fires when its inputs are ready or confirmed. Jev pre-suggests choices with a "we think this one" animation, still provision-first.

<!-- rules:stamp-end -->

Performance guarantee

The entire debug system is zero cost for non-admins:

• AdminIndicatorWrapper checks selectIsAdmin from Redux and returns null immediately if false
• Both AdminIndicatorIsland (ssr layout) and DynamicAdminIndicatorWrapper (authenticated layout) use dynamic(..., { ssr: false }) — the admin bundle never downloads for non-admins
• useDebugContext is a no-op when isAdmin is false — no dispatch, no effect, no render cost

---

Architecture — four layers

Layout (server)
  └── AdminIndicatorWrapper (client, admin-only)
        ├── AdminDebugContextCollector   ← auto: pathname, searchParams, viewport,
        │                                        console errors, unhandledrejection
        └── AdminIndicator (Small/Medium/Large)
              ├── MediumIndicator
              │     └── debugModules icons → DebugModulePanel modal
              └── LargeIndicator
                    ├── API config + health grid
                    ├── Recent API calls log
                    ├── [Copy Full Context] button ← assembles everything for AI agent
                    ├── Console Errors panel        ← captured automatically
                    └── Debug Data (JSON)           ← written via useDebugContext

---

What is auto-captured (no route work needed)

AdminDebugContextCollector runs at layout level inside AdminIndicatorWrapper. For every admin session it automatically captures:

Data	How
Current pathname	usePathname() on every navigation
Search params	useSearchParams() on every navigation
Browser viewport	window.innerWidth/Height
User agent	navigator.userAgent
Navigation count	increments on each pathname change
console.error calls	patches console.error while mounted
unhandledrejection events	window.addEventListener
window.error events	window.addEventListener

This is available instantly in the "Copy Full Context" snapshot even for routes that add zero custom debug code.

---

Adding route/feature debug data

The hook — use this everywhere

import { useDebugContext } from '@/hooks/useDebugContext';

export default function MyComponent({ sessionId }: { sessionId: string }) {
    const session = useAppSelector(s => selectSession(s, sessionId));
    const { publish, isActive } = useDebugContext('Chat');  // namespace

    useEffect(() => {
        publish({
            'Session ID': sessionId,
            'Status': session?.status,
            'Message Count': session?.messages.length,
            'Conversation ID': session?.conversationId,
        });
    }, [isActive, session?.status, session?.messages.length, sessionId]);
}

Key rules:

• First arg is the namespace — appears as group header in the debug panel
• Keys are auto-prefixed: "Chat:Session ID", "Chat:Status", etc.
• isActive is true only when isAdmin && isDebugMode — use it in useEffect deps to gate expensive collection
• The hook auto-clears all namespace keys on unmount — no manual cleanup needed
• Safe in any component tree — pure no-op for non-admins

Hook API

const { publish, publishKey, isActive } = useDebugContext('Namespace');

publish({ 'Key': value, 'Other Key': otherValue });  // merge namespaced data
publishKey('Key', value);                             // set single key
isActive  // boolean — true only when admin AND debug mode is on

---

Adding a debug module panel

For richer feature-specific UIs (not just JSON data), create a debug module:

Step 1 — Create the component

// components/admin/debug/MyFeatureDebug.tsx
'use client';
import { useAppSelector } from '@/lib/redux/hooks';

export default function MyFeatureDebug() {
    // Read exclusively from Redux selectors — no local fetch, no useState for data
    const data = useAppSelector(selectSomething);
    return <div className="p-4 type-secondary font-mono">{JSON.stringify(data, null, 2)}</div>;
}

Step 2 — Register in the registry

// components/admin/debug/debugModuleRegistry.tsx
import MyFeatureDebug from './MyFeatureDebug';
import { Zap } from 'lucide-react';

export const debugModules: DebugModule[] = [
    // ... existing modules ...
    {
        id: 'my-feature',
        name: 'My Feature',
        icon: Zap,
        component: MyFeatureDebug,
        description: 'Live state of my feature',
        color: 'text-purple-400',
    },
];

The icon appears in the MediumIndicator debug row. Clicking opens the panel modal. Component only mounts when the panel is open — zero overhead when closed.

---

Copy Full Context — the AI agent workflow

The Copy button in LargeIndicator assembles a complete markdown snapshot:

# Admin Debug Context Snapshot
Generated: 2026-03-27T10:42:00.000Z
Admin: admin@example.com

## Route
- **Path:** /demos/chat/c/abc-123
- **Viewport:** 1440×900
- **Render Count:** 3

## API Config
- **Active Server:** localhost
- **Backend URL:** http://localhost:8000
- **Health:** healthy (42ms)

## Feature Debug Data
### Chat
- **Session ID:** sess-xyz
- **Status:** streaming
- **Message Count:** 7

## Recent API Calls
| Status | Method | Path | HTTP | Duration |
...

## Console Errors
No errors captured.

Workflow: Admin sees a bug → opens LargeIndicator → clicks Copy → pastes into AI agent chat. Agent gets full context: what page, what state, what the API is doing, any errors.

---

Data sources reference

What	Selector	Slice
Is admin	selectIsAdmin	userSlice
Debug mode on	selectIsDebugMode	adminDebugSlice
Route context	selectRouteContext	adminDebugSlice
Console errors	selectConsoleErrors	adminDebugSlice
Debug data (JSON)	selectDebugData	adminDebugSlice
Active server	selectActiveServer	apiConfigSlice
Backend URL	selectResolvedBaseUrl	apiConfigSlice
Server health	selectActiveServerHealth	apiConfigSlice
Recent API calls	selectRecentApiCalls	apiConfigSlice
Active chat session	selectActiveChatSessionId	activeChatSlice
Chat session object	selectSession(state, sessionId)	chatConversations
Chat messages	selectMessages(state, sessionId)	chatConversations

---

Viewing debug context

1. Small indicator — two colored dots (server env, health)
2. Medium indicator — debug icon row; click any icon for that feature's panel
3. Large indicator — full admin dashboard:
   • Copy button (always visible) — one click to copy everything for an AI agent
   • Console Errors panel — auto-captured, collapsible, shows stack traces
   • Debug Data section (debug mode on) — namespaced JSON from useDebugContext
   • API calls log and all environments health grid

Enable debug mode: click the Debug toggle in MediumIndicator or LargeIndicator.