sessions-dashboard（会话仪表板）

Sessions Dashboard ("Live Dashboard Viewer")

Creator: Nate Teshome — <https://stellarsitesai.com>

What this skill delivers

• Color-coded CLI monitor (scripts/agents_cli_monitor.py): RUN/IDLE/STALE/EXIT states, runtime, last update, token + USD estimates, model/kind, plus a live KPI banner (state counts, Σ tokens/cost, longest run, avg lag).
• Shareable HTML dashboard (scripts/agents_canvas_snapshot.py): dark-mode cards + table, status pills, runtime/lag columns, and auto-refresh browser view for leadership/client reviews.
• Looping web generator (--loop <seconds>): continuously rewrites the HTML so you don’t need separate watch commands. Pair with the built-in <meta refresh> (default 5 s) for a true live panel.

Use this when you need a polished, always-on view of every OpenClaw session—perfect for ops rooms, agency war rooms, or lightweight audits.

Requirements & security

• OpenClaw CLI (openclaw >= 2025.x) must be installed and authenticated so openclaw sessions --json works.
• Optional log tail: the CLI monitor subscribes to openclaw logs --json --follow for instant RUN/IDLE bumps. Those logs can contain prompts, tool arguments, secrets, or PII. Nothing is printed or stored, but if you don’t want to expose log contents, pass --no-subscribe (poll-only) or run on a machine without log access.

- Set AGENT_MONITOR_NO_SUBSCRIBE=1 to default to poll-only everywhere; pass --subscribe to override per run.

• Gateway scope: run the dashboard only on hosts that are already allowed to query your OpenClaw gateway; everything stays local/in-memory.

Directory

skills/sessions-dashboard/
├── SKILL.md
├── scripts/
│   ├── agents_cli_monitor.py        # Live terminal UI (color + KPI banner)
│   └── agents_canvas_snapshot.py    # HTML generator (+ --loop for continuous refresh)
└── assets/
    └── agents_canvas.html           # Default output target for the web panel

CLI monitor quick start

cd ~/.openclaw/workspace/skills/sessions-dashboard
./scripts/agents_cli_monitor.py --interval 2 --cost-per-1k 0.015

Key flags:

• --interval SECONDS (default 3, min 0.5)
• --active-minutes MINS (default 720) to widen the lookback
• --agent <id> / --all-agents to filter
• --retention SECONDS (default 120) to control how long EXITED rows linger
• --cost-per-1k USD or env AGENT_MONITOR_COST_PER_1K
• --once for a single snapshot, --no-subscribe to disable the log tailer

What the CLI shows

• State (RUN green, IDLE amber, STALE magenta, EXIT gray)
• Runtime + start time for quick “how long has this been around?” checks
• Last update lag (00:12 ago) to spot stuck sessions
• Tokens & projected cost per session
• Model / kind (channel) for context
• KPI banner (top line) summarizing RUN/IDLE/STALE/EXIT counts, Σ tokens/cost, the longest-running session, and mean lag — ideal for screenshots or lightweight reporting.

Web dashboard (Canvas / browser)

Generate the HTML and keep it running in one command:

cd ~/.openclaw/workspace/skills/sessions-dashboard
./scripts/agents_canvas_snapshot.py \
  --loop 5 \
  --refresh 5 \
  --cost-per-1k 0.015 \
  --output assets/agents_canvas.html

What the script does:

• Hits openclaw sessions --json on each loop.
• Rebuilds assets/agents_canvas.html with cards (state counts, Σ tokens/cost, longest run, avg lag) plus the detailed table.
• Inserts a <meta refresh> so the browser (or Canvas tab) reloads itself every --refresh seconds.
• When --loop > 0, the script stays alive forever and rewrites the file at that cadence (Ctrl+C to stop). No extra watch command needed.

Presenting the dashboard

• Open locally: open assets/agents_canvas.html
• Canvas: openclaw canvas present --path skills/sessions-dashboard/assets/agents_canvas.html
• Remote share: host the HTML anywhere static files are allowed; it has no external dependencies.

Troubleshooting

• command not found: chmod +x scripts/*.py and run from the skill directory (or use absolute paths).
• Gateway errors: verify openclaw gateway status; the scripts just print the stderr and retry.
• No color in CLI: set NO_COLOR=1 to disable, otherwise the monitor auto-detects TTY support.
• Browser not updating: make sure --loop is running so the HTML file actually changes; --refresh only reloads the page.
• Cost column blank: pass --cost-per-1k or export AGENT_MONITOR_COST_PER_1K to enable projections.

Release notes (current version)

• Added colorized states + KPI banner to the CLI monitor.
• Added looping mode and a polished dark-mode layout for the HTML dashboard.
• Default assets are optimized for screenshots / executive readouts.

That’s it—launch the CLI, fire up the live HTML panel, and give your team a real-time Sessions Dashboard without cobbling together spreadsheets.