Pipeline & History

Pipeline panel

The main window’s “All Projects” view and each per-project Pipeline. A live cross-project feed of staged, pending, running, completed, and failed prompts with approve, cancel, retry, edit-and-approve, and per-project queues.

MCP tools: orbit_status, orbit_status_markdown.

Completions Log & raw logs

Project window → Completions tab, with a “View Raw Log” modal. An append-only per-project COMPLETION_LOG.md audit trail (prompt #, description, commit, model, notes); every run also writes a raw stdout/stderr transcript under logs/prompts/, viewable inline.

MCP tools: orbit_get_completions, orbit_get_completion_full, orbit_get_prompt_log.

Visual-evidence contract

A prompt that changed the UI is expected to prove it — a screenshot, a recorded interaction, or an explicit stated reason the change couldn’t be visually verified. When a shipped UI-changing prompt’s completion carries none of these, orbit_get_completions surfaces a missing_visual_evidence flag on that entry so it’s visible in the Completions tab rather than silently trusted. Backend-only work is never held to this — the check only fires when the diff actually touched UI.

Needs You

Dashboard health banner and Pipeline. Needs You is where Orbit surfaces things that need a human look rather than folding them silently into a status color: canonical checkout drift across configured projects (isolating unhealthy peers from healthy ones), a Needs Review card for a run whose status is ambiguous (mark it shipped, re-queue it, or dismiss it, with inline Git evidence backing each option), and dependency waits that couldn’t be confirmed automatically and fail safe into a manual-review note rather than silently blocking or silently proceeding. Every admitted Needs You item can be dismissed once you’ve reviewed it. Prompt-owned outcomes — push, partial, and verification results — stay on their own Pipeline cards with truthful lower-bound coverage rather than being absorbed into the Needs You banner.

A related Failed-ship audit re-checks recent terminal failed rows whose work is provably reachable on the canonical branch despite the recorded failure, and flags them for review without silently rewriting the original verdict.

Integration and publication status

Pipeline and Completions rows. A finished prompt’s terminal outcome is settled in one transaction — a remotely confirmed integrated commit (or an equivalent applied patch) updates durable push state and canonical Shipped status together, so failed publication and merge conflicts stay explicit instead of quietly resolving themselves. Completions rows also show whether every commit owned that prompt actually reached the integration target by content, distinguishing proven-missing work from a replay that was merely inconclusive. If a dependency-integrity check trips, queueing and read-only inspection stay available while it’s tripped; it re-arms automatically once inspections come back clean, or an operator can reset it after fresh verification (orbit_rearm_dependency_guard).

See also Git Tab & Recovery for the related Git-side recovery surfaces (stash inventory, dirty-pull recovery, unreachable-commit warnings, and publication repair doors).

History sidebar (Activity panel)

The right-rail History/Activity panel.

Period selector — a Day / Week / Custom control sets the window for all metrics and the chart. Custom lets you pick an arbitrary date range.

Hero stat and chart — the selected period’s total event count as a headline number, backed by a multi-line SVG chart showing the overall trend plus individual per-project lines with gradient fill.

Project ranking and toggles — projects are ranked by activity for the period. Each project row has plot toggles so you can show or hide individual projects from the multi-line chart without changing the ranking.

Kind filter — a tab strip filters the activity feed to All, Prompts, QA, or Deploys.

Weekly record badge — when the current week’s count matches or exceeds the all-time weekly record, a record badge (“Record: N”) appears next to the hero stat.

Exports — a range-export action produces a file in your choice of format: CSV (columns: timestamp, kind, project, title, model, status, meta), JSON (full event objects), or Markdown (a summary table of prompt/QA/deploy counts per project). Downloads go directly to the browser/filesystem.

Daily Summary modal

The “Daily Sum” button in the All-Projects pipeline header. A three-tab recap of the calendar day — Numbers (KPI grid and sparklines), Story (an AI-generated narrative), and Timeline (a per-project activity heatmap).

Notifications bell

The dashboard titlebar bell. An actionable-by-default feed — failures, stuck or ghost prompts, deploy and QA failures, queue-paused events — with project and prompt numbers, plus a “Show all events” toggle for the full feed.

Diagnostics panel & metrics

Ctrl+Shift+D. Live System Load, Project Growth, and Largest Dirs metrics with anomaly tinting (such as worktree-count alerts), section copy and JSON export, and resizable docking.

Pipeline scroll flight recorder — Diagnostics → Pipeline Scroll Trace. Captures a bounded, correlated trace of Pipeline scrolling alongside renderer and server activity, with incident markers, rotating trace files, and a ranked likely-cause diagnosis — for tracking down fast-scroll or stutter failures that are otherwise hard to reproduce on demand.

Renderer crash logger

Unhandled renderer errors and rejections are written to ~/.orbit/renderer-crash.log and replayed in a startup banner — no DevTools needed.

Audit Center

The sidebar Audits entry and the Docs tab’s audits/ subfolder. A severity-banded list with keyboard triage of diagnostic audit reports, each with a structured detail view (verdict, flaws, “Orbit’s proposed fix”) and a one-click “Queue this fix.”