Skip to content

README-DASHBOARD.md

Latest commit

 

History

History
149 lines (107 loc) · 5.11 KB

README-DASHBOARD.md

File metadata and controls

149 lines (107 loc) · 5.11 KB
Raw

Agent Dashboard

Web-based monitoring and control interface for Product Engineer agents.

Quick Start

Visit: https://product-engineer.<your-subdomain>.workers.dev/dashboard

Login with your Google account (must be on the allowlist).

What You Can Do

Monitor Agents

  • View all active agents in real-time
  • See ticket IDs, products, status, and latest activity
  • Agents needing help (asking status or stale heartbeats) shown at top
  • Direct links to Slack threads and GitHub PRs
  • Auto-refreshes every 30 seconds

Control Agents

  • Kill individual agent: Stop a specific agent immediately
  • Kill all agents: Emergency shutdown of all active work

Security

  • Google OAuth: Secure authentication flow
  • Email allowlist: Only authorized users can access (configured via ALLOWED_EMAILS secret)
  • Session cookies: 24-hour lifetime, HttpOnly, Secure
  • CSRF protection: State parameter validation

Setup Instructions

See docs/dashboard-setup.md for complete deployment guide.

Quick Deploy

# 1. Create KV namespace
cd api
wrangler kv namespace create SESSIONS

# 2. Update wrangler.toml with namespace ID

# 3. Configure Google OAuth in Cloud Console

# 4. Set secrets
wrangler secret put GOOGLE_CLIENT_ID
wrangler secret put GOOGLE_CLIENT_SECRET
wrangler secret put ALLOWED_EMAILS  # Comma-separated: user1@example.com,user2@example.com

# 5. Deploy
wrangler deploy

API Reference

All endpoints require authentication.

Get Agents

GET /api/dashboard/agents

Returns list of active agents with status, heartbeats, and thread info.

Kill Agent

POST /api/dashboard/agents/:ticketId/kill

Marks agent inactive and requests container shutdown.

Kill All

POST /api/dashboard/agents/shutdown-all

Emergency shutdown of all active agents.

UI Overview

┌─────────────────────────────────────────────┐
│ 🤖 Agent Dashboard        👤 user@email.com │
├─────────────────────────────────────────────┤
│ [🔄 Refresh] [🛑 Kill All]  Auto-refresh: 30s│
├─────────────────────────────────────────────┤
│ ⚠️ Agents Needing Help (2)                  │
│ ┌─────────────────────────────────────────┐ │
│ │ Ticket  Product  Status  Thread  Heartbeat│
│ │ BC-123  bike     asking  [Link]  5m ago  │ │
│ │ HT-45   health   in_prog [Link]  35m ago │ │
│ └─────────────────────────────────────────┘ │
├─────────────────────────────────────────────┤
│ ✅ Active Agents (5)                         │
│ ┌─────────────────────────────────────────┐ │
│ │ Ticket  Product  Status   Thread  Updated│
│ │ BC-124  bike     pr_open  [Link]  2m ago │ │
│ │ HT-46   health   in_prog  [Link]  1m ago │ │
│ │ ...                                      │ │
│ └─────────────────────────────────────────┘ │
└─────────────────────────────────────────────┘

Troubleshooting

"Access denied" → Your email is not on the allowlist. Contact admin to add you.

"Session expired" → Sessions last 24h. Just refresh to log in again.

Dashboard shows no agents → Agents may be inactive. Check orchestrator status API.

Kill doesn't work → Check orchestrator and agent logs. Container may already be stopped.

Development

Dashboard is a single-page app (src/dashboard.html). No build step needed.

Auth logic in src/auth.ts. Routes in src/index.ts.

Local dev requires public tunnel for OAuth redirect (ngrok, cloudflared).

Monitoring Use Cases

Daily Operations

  • Check agent health at start of day
  • Identify stuck agents (stale heartbeats)
  • Monitor asking agents (need user input)

Incident Response

  • Kill runaway agents consuming resources
  • Emergency shutdown during deployment issues
  • Verify agents stopped after maintenance

Capacity Planning

  • Track typical agent count
  • Identify busy times
  • Plan container limits

Future Enhancements

  • Email allowlist UI (admin can add/remove users via dashboard)
  • Audit log for kill operations
  • Agent logs viewer (tail agent container output)
  • Historical stats (agents per day, success rate)
  • Slack notifications for dashboard kills
  • Agent resource usage (CPU, memory, duration)
  • Batch operations (kill multiple selected agents)
  • Search/filter agents (by product, status, time range)