Buttons
Pills (status — what something is DOING)
running
thinking
needs approval
idle
errored
K-badges (kind — what something IS)
tools
approval
policy
role
judge
query
workstream
session
Chips + risks
ws_a1b2c3d4
tok_f00bar
node-04
low
med
high
crit
Typography scale
t-h1 — heading (24px)
42,195
t-kicker — section label
t-body — base 14px / 1.45 for prose.
t-row — 13px for table rows and sidebar items.
t-btn — 12px for button labels.
Stat row
Workstreams
28
+3 / 24h
Running
14
steady
Thinking
3
-1
Attention
2
+2
Idle
8
—
Errored
1
+1
Panels + mini-bar + spark
Load
CPU
Memory
Fleet
ok
thinking
attention
error
idle
unreachable
Live feed
10:42:18
tools
spawn_batch
workstream ws_a1b2c3d4
Created 6 child workstreams to fan out log-triage across shards.
{"shard": "eu-1", "count": 6, "budget_remaining": 18}
0.94
10:42:21
approval
exec
node-12 / rm -rf /tmp/cache
Requesting approval to clear cache on node-12 — low risk.
10:41:02
policy
allow
role:operator
Applied policy bundle v4 (3 deny rules, 2 step-up requirements).
judging
Live feed (pattern)
10:42:18
tools
spawn_batch
ws_a1b2c3d4
Fanning out 6 child workstreams to shard log triage.
0.94
10:42:19
judge
score
child ws_ff11
Scored child output: 0.88 confidence.
0.88
10:42:20
policy
allow
role:operator
Policy bundle v4 applied.
judging
Messages (chat primitive)
Can you fan out a log-triage across the three eu shards and tell me which one had the most 5xx in the last hour?
Spawning six child workstreams (two per shard) to grep the last hour's access logs. Will aggregate the status >= 500 counts and report back.
{"shards": ["eu-1", "eu-2", "eu-3"], "lookback": "1h", "filter": "5xx"}
Let me think about how to partition this — per-shard, or per-node inside each shard?
spawn_batch → 6 children created, budget: 18 remaining
Policy bundle v4 applied to this session.
node-04 did not respond within 5s; retrying with backoff.
Session ended by operator.
Form fields (primitive)
Appbar (chat header chrome)
Semantic token swatches
accent
accent-soft
ok
warn
err
think