Panels

Interface reference

The observatory’s panels, named and annotated — so documentation, issues, and support can refer to each view precisely. Screens are from VS Code; JetBrains mirrors them.

01

The workspace

Every region and toolbar, called out on one screen: the Claude Edits sidebar and the editor on top, the Claude Observatory bottom panel below, and the status bar along the bottom.

02

The panel views in detail

The observatory’s own views, with every part named: the Observations · Overview · Stats panes of the “Claude Observatory” bottom panel, plus Actions, which lives in the Claude Edits sidebar.

Overview panel, master-detail. LEFT NAV (~25%), three tabs, each opening with a one-line description of what it shows (tabs and change-map sections also carry hover descriptions): 1 Fleet — every running agent across the repo's git worktrees, each with a live phase badge (working / awaiting-input / awaiting-permission / idle / errored / done; heuristic phases carry a ~ marker — inferred from inactivity), worktree + branch, an activity sparkline, ±diff, tokens·time, risk and a collision badge, unfolding to nested subagents (agentType/description, phase, current task + to-dos, ±lines, a chat button); 2 Workflows — each multi-agent workflow run, with an informative name, running/done, per-phase progress groups and its agents (tokens·time·edits) over a matching sparkline; 2b Tasks — the session's numbered task list (Claude's TaskCreate/TaskUpdate tasks) with live statuses, a done badge, and strikethrough on completion; plus an Active-only toggle (which also hides completed tasks) and Clear-completed (dismisses, never deletes). RIGHT DETAIL (~75%): 3 the selected agent's change-map, in three labeled sections top to bottom — Chapters, a ribbon of subtask chips (each a to-do/task Claude worked, showing its lines / edits / pending; chapters are total — work outside any to-do appears as a dimmed synthesized session chapter titled from the session's goal, reviewable like any other; duplicate to-do titles render as separate rows); Folders, a proportional strip of tiles, one per changed folder (tile width = lines changed, color = review status); and Files, a churn-ranked ledger of every changed file (±line bars colored by review status). Clicking a Folder tile or a Chapter chip navigates the matching nav-bar axis, opening that folder's or chapter's first pending edit. A bottom summary bar reports the pending / accepted / reverted edit counts plus file and folder totals for whatever is currently in scope, and names the current chapter (or folder filter). NAV BAR: 4 the review toolbar, now two rows. Top row (controls): the session selector — showing the session's human-readable name, its title or first prompt, with the raw id in the tooltip — then the session-wide bulk actions Accept All · Revert All · Clear Resolved · Export, and on the right Search · Active only · Spotlight · Refresh. Bottom row: four review axes that step the pending edits at four granularities — Diff axis steps the open file's edits, carrying Keep, Undo, Chat (hands this edit's context to your Claude), View diff (opens a real side-by-side diff editor) and the current edit's relative time; File axis steps changed files, showing the filename and that file's edit count, with Accept/Reject File; Folder axis steps changed folders, showing the folder and its file/edit totals, with Accept/Reject Folder acting on that folder's edits only; Chapter axis steps the session's subtasks (chapters, from Claude's to-dos/tasks), showing the chapter's folder/file/edit totals, with Review, Accept Chapter, Reject Chapter and Chat — labeled, color-coded, one icon per action — Phases marked ~ are inferred from inactivity, never asserted. The nav bar carries the session controls on its top row and four review axes — Diff, File, Folder, Chapter — on the bottom; the change-map’s Chapters, Folders, and Files sections mirror them, and clicking a folder tile or chapter chip drives the matching axis. The chapter ribbon is total — work outside any to-do lands in a dimmed session chapter, reviewable like any other.

The Overview's Fleet left-nav tab (formerly the Multitasking window): 1 per-agent rows — each running agent across the repo's git worktrees, with a live phase badge (heuristic phases carry a ~ marker), worktree + branch, an activity sparkline, ±diff, tokens·time, risk and a collision badge; 2 nested subagents (agentType/description, phase, current task + to-dos, ±lines, a chat button); 3 live cross-agent file conflicts (shown in the Actions view)

The Overview's Workflows left-nav tab: one row per multi-agent workflow run — an informative name, a running/done badge, phase progress with tokens and elapsed time, and an activity sparkline — unfolding to per-phase groups and their agents (tokens · time · edits). Selecting a run drives the right-hand change-map detail with that run's chapter rollup and files — Each workflow run maps its agents’ edits into the same change-map detail as any fleet agent.

The Overview's Tasks left-nav tab: the session's numbered TaskCreate/TaskUpdate tasks, newest first — a status glyph (in-progress / done with strikethrough), the task number and subject, the in-progress task's active form in italics, and per-task +added −removed and edit counts joined from the task's chapter. Completed tasks fold behind an 'N done · show all' row, and the Active-only toggle hides them outright — Tasks link to chapters: each task’s row carries the live ± and edit counts of the chapter it planned, and reviewing that chapter reviews the task’s edits.

Observations panel: 1 session recap, 2 the coalesced change feed (files ordered by recent activity; adjacent same-file edits collapse into a ×N run) — the former Timeline is folded in here, 3 per-edit reasoning rows with flags and file-memory, 4 next steps

Stats panel: 1 review scoreboard (the session shown by name, not its id), 2 range toggle, 3 token plot, 4 usage bars — the Usage 5h and weekly rows now show an estimated used / total, the 100% total inferred from tokens ÷ percent

Actions view — now in the Claude Edits sidebar, not the bottom panel: every tool call Claude made, grouped by category (Edits · Commands · Reads · Searches · Web · To-dos · Egress) and collapsed by default — led by a Live-conflicts section (files touched by 2+ active agents, click to open) that starts expanded; errors always surface, a Show-all toggle reveals the low-signal categories, and edit rows link to their review. The Risk and Egress audits ride the same timeline

Interface reference

The workspace

The panel views in detail

Overview master-detail · the fleet and the change-map

Overview → Fleet every running agent, live

Overview → Workflows multi-agent orchestration runs

Overview → Tasks the session’s numbered task list

Observations the change feed · Claude's reasoning

Stats progress & usage

Actions every tool call — in the sidebar