Deck v2 — updated to your decisions

Overview is home (scoreboard + urgent inbox + widgets) · Inbox stays a full page · Map = Atlas page · run-forensics widgets from the harness report (built once, shared with the dynwf-ux plan endpoints) · Botmaster widget added · dark default with light mode (toggle bottom-left, preview at the bottom) · TUI keeps direction C, built from shared ratatui components.

Overview Inbox5 Map
Operations
Plans1▶ 1⛔ Agents8 CI & Runners1 Bots4
Health
Limits & Spend Gates Scoreboard
☀ Light◐ Dark

Overview

laptop · debian1 · harness:4973
Search or command… ⌘K

multideal scoreboard →

8 /34▲ +2 wk
M4 prod env next · CI red (e2e)

zync.is

3 /31■ flat 9d · 118 commits — churn
Z2 dead pay-links still open

press.zone

6 /10▲ +1 · customer date Jul 25
Hetzner prod · nightly smoke green

Needs you now open inbox (5) →

Orphan agent burning 94% CPU for 3h12m
cursor-agent 88231 · ignores SIGTERM · agent-guard
multideal CI #4821 failed — e2e checkout spec
laptop runner · 2nd failure today, same spec
Plan HALT: dep-provisioning needs a decision
wave 3 · "pin drizzle 0.44 or bump?" · waiting 42m
Claude account A at 78% — caps in ~55m
systray-ai forecast · account B at 23%

Limits & spend

claude A78% · ~55m left
claude B23%
codex41%
grok12%
spend today$11.40
Run forensics — autonomy-p4-babysitter · last engine run · shared component with runplan TUIall runs →
30.2h
wall clock
17 tasks, 16 landed
9
process runs
3 crashed, no terminal event
76%
time in LLM seats
fixer 11.3h · impl 7.6h · review 3.7h
218m
longest hang
configured cap: 60m — never killed
13
intervention types
babysitter LLM performs by hand
90%
disk full
91G free · debris census below

Where the time went

Fixer (LLM)
11.3h · 37%
Implement (LLM)
7.6h · 25%
gate0 build+test
4.8h · 16%
Review (LLM)
3.8h · 12%
Idle between runs
2.0h · 7%

Runs full timeline →

run #3 · 9.05h · landed 14/17
run #4 · 4.46h · crashed
run #5 · 4.19h · crashed
run #6 · 3.48h · landed 15/17

Bots (Botmaster) bots →

BotStatusMsgs 24hErrorsCost
dealbotlive4120$1.9
supportbotlive961$0.7
alertbotslow2304$0.3
zyncbotdown 2h0

CI & Runners

multideal #4821✗ e2e · 14m
platform #312✓ 6m
laptop runnerbusy
debian1 rb-*idle · online
queue0 waiting

Plans

autonomy-p4wave 2/5 ▶
dep-provisioningHALT 42m
golive-nightly03:00 · debian1
agents live8 · 1 orphan

Hosts & Gates

PSI cpu some4.1%
mem runway> 6h
PROCHOT0
prevent-band0 open
slopgate debt14 ▼2

Light mode (same tokens)

8 /34▲ +2 wk
Theme = ui-tokens theme-engine · toggle or system
Every widget above is a standalone component (KpiTile, PhaseBars, RunLanes, ScoreCard, InboxItem, LimitMeter, BotTable, KvPanel) — same components power the full pages (Inbox, Map, Plans drill-down) and the run-forensics set is the SAME component the runplan TUI's web twin + reports use. Nothing rendered twice.
Reply in terminal: approve, or name changes. TUI stays direction C (approved) — built from a shared ratatui component kit extracted from the existing runplan TUI.