OpenClaw Dashboard

OpenClaw operations dashboard for sessions, usage and cost, cron runs, gateway health, DGX Spark work, Local API Hub, and opt-in meeting Copilot. Use when installing, operating, au…

Jonathan Jing

@jonathanjing

What This Skill Does

Real-time operations dashboard for OpenClaw that monitors sessions, costs, cron jobs, and gateway health. Includes a language toggle (EN/中文), watchdog 24h uptime bar, and cost analysis.

Replaces manually checking multiple OpenClaw logs and metrics by consolidating all operational data into a single web dashboard.

When to Use It

  • Monitor active OpenClaw sessions and their associated costs
  • View cron job execution history and status
  • Check gateway health and uptime via the watchdog bar
  • Toggle dashboard language between English and Chinese
  • Add new backend routes to the api-server.js file
  • Update frontend behavior in agent-dashboard.html

Install

$ openclaw skills install @jonathanjing/openclaw-dashboard

OpenClaw Dashboard Skill

Preview

The published screenshots use deterministic ?preview=1 sample data, never local host or workspace data.

OpenClaw Dashboard overview

OpenClaw Dashboard usage analytics

<img src="https://raw.githubusercontent.com/JonathanJing/openclaw-dashboard/71604d6831bb24b9e8340bf0d1153a9545c5a038/screenshots/mobile-v2-dark.png" alt="OpenClaw Dashboard mobile layout" width="390">

🛠️ Installation

1. Ask OpenClaw (Recommended)

Tell OpenClaw: "Install the openclaw-dashboard skill." The agent will handle the installation and configuration automatically.

2. Manual Installation (CLI)

openclaw skills install @jonathanjing/openclaw-dashboard

Mission

Keep this repository public-safe and easy to run. Prioritize:

  1. Secret sanitization
  2. Minimal setup steps
  3. Stable API/UI behavior

Architecture (v2.0)

The dashboard uses a modular backend + tab-based frontend architecture.

Backend entry point: backend/server.js
Business logic: backend/providers/ — one file per data domain
Frontend: frontend/tabs/ + frontend/shared/ — one JS file per tab
Runtime data: stored in ~/.openclaw/dashboard/ (outside skill dir, not Git-tracked)

Provider map

ProviderRoutesResponsibility
sessions.js/ops/sessions, /api/sessionsSession stats + model
ledger.js/ops/ledger/*, /api/cost/*SQLite token/cost data
cron.js/ops/cron, /ops/cron-costs, /cron/todayCron jobs + run history + cost breakdown
watchdog.js/ops/watchdogWatchdog state + timeline
spark.js/ops/dgx-status, /api/spark/*DGX Spark inference node
system.js/ops/systemHost metrics (CPU/RAM/disk)
local-api-hub.js/ops/local-api-hub/*Unified local control-plane status
spark-tasks.js/api/spark/tasks/*DGX task history and PR Hunter output
copilot.js/api/copilot/status, /api/copilot/wsOpt-in realtime transcript/RAG/insights
ground-truth.js/api/ground-truth/*, /ops/modelsModel registry + colors
tasks.js/tasks, /tasks/:id, /logsRead-only task/history views
config.js/ops/config, /files, /skillsRead-only config, file, and skill views
ops-legacy.js/ops/channels, /ops/alltime, /ops/audit, /memoryRead-only compatibility views

Frontend tab map

TabFileKey functions
Overviewtabs/overview.jsloadSessions(), loadTasks()
Costtabs/cost.jsloadOpsChannels(), loadOpsAlltime()
Crontabs/cron.jsloadCronEnhanced(), loadCronCosts(), loadCronRuns()
Healthtabs/health.jsrenderAgentMonitor(), loadSystemInfo(), renderWatchdogStatus()
Sparktabs/spark-monitor.jsloadSparkMonitor(), task history, GPU activity
Copilottabs/copilot.jscapability check, microphone stream, transcript/RAG/insights
Configtabs/config.jsloadConfig(), loadSkills(), loadFileList()
Sharedshared/api.jsAuth, apiFetch(), watchdog renderers, toast, markdown
Sharedshared/ui-utils.jstimeSince(), task state
Sharedshared/boot.jsInit, week nav, chart renderers, confirm dialog

Apply when

Use this skill for:

  • Dashboard feature requests (sessions, cost, cron, watchdog, operations)
  • Backend route additions/fixes in backend/providers/
  • Frontend behavior updates in frontend/tabs/ or frontend/shared/
  • README, setup, and environment simplification
  • Public release checks for accidental sensitive data

Key rules for agents editing this codebase

  1. No duplicate function definitions across api.js and ui-utils.js. Shared utilities belong in api.js (loaded first). ui-utils.js only holds timeSince() and task state.
  2. Cross-tab function calls are implicit — JS shares the same window scope. Keep shared helpers in shared/ files.
  3. Runtime data goes to ~/.openclaw/dashboard/, not skill root. Path is set in backend/lib/config.js via OPENCLAW_DASHBOARD_TASKS env or default.
  4. /ops/models returns { registry: {...object...}, colors, displayNames, models } — registry must be an object keyed by alias, not an array.
  5. /ops/cron-costs returns { summary, jobs, dailyTrend, review, rows } — all five keys required for Cron tab to render correctly.
  6. hideStale query param on /ops/sessions filters sessions with no activity for 7+ days.
  7. Frontend API calls are same-origin. Never reintroduce a hardcoded dashboard/gateway port list.
  8. Copilot is opt-in. Require OPENCLAW_ENABLE_COPILOT=1, an API key, authenticated WebSocket upgrade, and safe dependency failure states.
  9. Parse request URLs against a fixed internal base. Never build a URL from the request Host header.
  10. Treat query-token login as a compatibility handoff only. Set the cookie and redirect before serving HTML; never accept query tokens on API routes.
  11. Parse cookies per fragment. Split on the first =, catch percent-decoding errors, and encode cookie values when setting them.
  12. Keep the shipped dashboard read-only. Do not add task/file mutations, restart/doctor actions, model changes, package updates, backup/restore, or legacy proxying.
  13. Scope Copilot Redis events by meeting ID. Permit unscoped legacy channels only for the first active meeting.
  14. Track the launcher and test harness. Keep start.sh and scripts/test-dashboard.js in both Git and package.json#files.
  15. Use theme tokens in canvas rendering. Never hardcode light-only chart text or borders.

OpenClaw design alignment

  • Reuse the current Control UI tokens: Inter/system typography, layered neutral surfaces, thin borders, 10–14px radii, red primary accent, semantic green/yellow/red/blue.
  • Keep the desktop shell as left navigation + sticky topbar; use bottom navigation on narrow screens.
  • Prefer flat cards and strong information hierarchy over gradients, glow, or decorative motion.
  • Support dark and light modes, visible focus states, reduced motion, and responsive layouts.
  • Keep IDs and data-tab contracts stable when changing navigation or visual structure.

Public-safety guardrails

  • Never hardcode tokens, API keys, cookies, or host-specific secrets.
  • Require OPENCLAW_AUTH_TOKEN; do not start the dashboard without authentication.
  • Never commit machine-specific absolute paths.
  • Prefer process.env.* and safe defaults based on HOME.
  • Keep examples as placeholders (your_token_here, /path/to/...).
  • If uncertain, redact first and ask the user before exposing details.
  • Keep sensitive behaviors opt-in (do not silently load local secret files).

Runtime access declaration

The bundled server can access local OpenClaw files for dashboard views:

  • Sessions, cron runs, watchdog state under ~/.openclaw/...
  • Local workspace files under OPENCLAW_WORKSPACE
  • Task data in ~/.openclaw/dashboard/tasks.json
  • Task attachments in ~/.openclaw/dashboard/attachments/

High-sensitivity features are disabled by default and require explicit env flags:

  • OPENCLAW_ENABLE_CONFIG_ENDPOINT=1 to expose /ops/config
  • OPENCLAW_ENABLE_COPILOT=1 plus ALIBABA_CLOUD_API_KEY to enable meeting Copilot

Network security:

  • CORS restricted to loopback by default.
  • Auth via HttpOnly cookie (ds) or Authorization: Bearer header.
  • Set DASHBOARD_COOKIE_SECURE=1 only when the dashboard origin uses HTTPS.
  • Set OPENCLAW_CONTROL_UI_URL to the complete runtime URL when Gateway TLS or gateway.controlUi.basePath is enabled.
  • Set DASHBOARD_CORS_ORIGINS (comma-separated) for external origins.

Default implementation workflow

  1. Identify which provider or tab file owns the feature.
  2. Implement the smallest change that preserves behavior.
  3. Check: does any other tab/shared file also define the same function? If yes, deduplicate.
  4. Run a sensitive-string scan before finalizing.
  5. Run npm test, git diff --check, and npm pack --dry-run --json.
  6. Ensure docs match the actual runtime defaults.

Sensitive-data checks

Before final response, scan for:

  • token=, OPENCLAW_AUTH_TOKEN, OPENCLAW_HOOK_TOKEN
  • API_KEY, SECRET, PASSWORD, COOKIE
  • absolute paths like /Users/, C:\\, machine names, personal emails

If found: replace with env-based values or placeholders, and mention what was sanitized.

Files to touch most often

  • backend/providers/*.js — server behavior and API routes
  • frontend/tabs/*.js — tab-specific UI logic
  • frontend/shared/api.js — auth, fetch, shared renderers
  • backend/lib/config.js — path and env configuration
  • README.md — quick start and operator docs
  • env.example.txt — ClawHub-compatible public-safe environment template
  • env.example — compatibility mirror for Git/npm workflows

Top skills in this category

claude-code-usage

@azaidi94

Check Claude Code OAuth usage limits (session & weekly quotas). Use when user asks about Claude Code usage, remaining limits, rate limits, or how much Claude usage they have left. Includes automated monitoring to notify when quotas reset.

167.1k

Skill Vetter

@spclaudehome

Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.

1.3k278k

Free Ride - Unlimited free AI

@shaivpidadi

Manages free AI models from OpenRouter for OpenClaw. Automatically ranks models by quality, configures fallbacks for rate-limit handling, and updates opencla...

49172k

Playwright (Automation + MCP + Scraper)

@ivangdavila

Automates, tests, and debugs browsers with Playwright: locators, auto-waiting, traces, CI runs, and MCP browser control. Use when a test is flaky, times out, or fails only in CI or headless; when a locator matches multiple elements or the wrong one (strict mode violation); when clicks need force, waits become sleeps, or networkidle never settles; for storageState and login setup, request mocking and HAR replay, uploads and downloads, iframes and shadow DOM, popups and dialogs, screenshot diffs that change per machine, trace and report artifacts, sharding a slow suite, device and permission emulation, accessibility checks, driving a real browser through Playwright MCP, extracting data from JS-rendered pages, or porting a Cypress, Puppeteer, or Selenium suite to Playwright. Not for maintaining an existing Cypress or Puppeteer suite (cypress, puppeteer) or for work a plain HTTP request answers (http).

13043k

n8n

@thomasansems

Manage n8n workflows and automations via API. Use when working with n8n workflows, executions, or automation tasks - listing workflows, activating/deactivating, checking execution status, manually triggering workflows, or debugging automation issues.

6121k