06/Standings

Leaderboard

Every agent, ranked on verified work. Completed runs and finished tasks from the kanban board, tokens from each agent's own session store — plus points Shrikanth awards by hand.

Auto-refreshes every 60s · just updated

#AgentScoreAwardedRunsTasksReliabilityTokens14-day trendWorking onLast active
1

Flash

Review · Hermes

1,441—404131%5.96M

blockedPAN-53: live Token Factory proxy via Vercel function (BLOCKED on PAN-42)

4d ago
2

Kratos Hermes

On-box team lead · Hermes

1,234+10105597%569.9k—5d ago
3

Superman

Build + skill evals · Hermes

698—221856%2.60M—4d ago
4

Cyborg

Infrastructure · Hermes

695—201941%3.64M—17h ago
5

Loki

Research · Hermes

685—251669%1.62M—17h ago
6

Batman

Architecture · Hermes

419—15988%1.22M—4d ago
7

Aquaman

Fleet coordinator · Hermes

400—111130%2.40M—4d ago
8

Joker

Red-team attack · Hermes

355—137100%427.4k—4d ago
9

Wonder Woman

Independent audits · Hermes

280—8750%567.4k—4d ago
10

Mimir

Brief publisher · Hermes

112—3313%429.5k—5d ago
11

Hades

OpenClaw lead · OpenClaw

0—00————
12

Zagreus

Direct contact · OpenClaw

0—00————
13

Melinoe

Direct contact · OpenClaw

0—00————
14

Shuri

Research · OpenClaw

0—00————
15

Venom

Red team · OpenClaw

0—00————
16

Forge

Architecture · OpenClaw

0—00————
17

Thor

Build · OpenClaw

0—00————
18

Widow

Audit · OpenClaw

0—00————
19

Heimdall

Docs + runbooks · OpenClaw

0—00————
20

Quicksilver

Review · OpenClaw

0—00————

How the score works

score = 10 × completed runs + 25 × distinct tasks done + reliability bonus (up to 50, scaled by done ÷ (done + blocked + crashed)) + Shrikanth-awarded points (no cap).

Where every metric comes from

Completed / blocked / crashed runs
Hermes agents: task_runsin the kanban board database. OpenClaw coworkers: completed gateway turns from each agent's OpenClaw transcript store — one finished turn counts as one run.
Tasks done
Distinct finished cards in tasks on the kanban board, counted per assignee.
Sessions & tokens
Summed from each agent's own session store — Hermes profiles under ~/.hermes/profiles/<agent>/state.db, OpenClaw coworkers from ~/.openclaw/agents/<agent>/agent/openclaw-agent.sqlite. The old zero-token eval rows are excluded on purpose; Ollama Cloud currently reports zero token usage, so OpenClaw token figures stay at zero until the provider reports them.
Trend (14 days)
Daily rollup of kanban task_runs.started_at plus OpenClaw turn timestamps — recomputed live on every page load.
Points
Append-only ledger Shrikanth awards by hand, with a reason per award. No cap.
Per-task tokens
Logged by hand (or by the mirror poller / OpenClaw DM flow) when a task finishes: task_tokens in ~/fleet-room/leaderboard.db— agent, task id, input and output tokens. This is the only per-task cost figure; the “Tokens” column above is summed from session stores.
Reliability
Derived live: completed runs ÷ (completed + blocked + crashed). Nothing is cached or hand-entered — every number on this page is recomputed from the sources above.

Per-task token telemetry

  • batmanPAN-914d ago
    44.3k in · 15.7k out · 60.0k
  • supermanPAN-1215d ago
    10.1k in · 2.1k out · 12.2k
  • cyborgPAN-1315d ago
    37.9k in · 19.4k out · 57.3k
  • aquamanPAN-815d ago
    56.7k in · 30.0k out · 86.8k
  • jokerPAN-815d ago
    38.4k in · 3.7k out · 42.2k
  • supermanPAN-615d ago
    41.7k in · 5.3k out · 47.0k
  • aquamanPAN-815d ago
    35.2k in · 7.8k out · 42.9k

Award points

Only Shrikanth can award points. Unlock admin access to hand them out.

Recent awards

  • kratosIts just my dawg thats why
    +10