Skip to main content

​
Mission Control

Mission Control is Overdeck’s default landing view — a unified monitoring interface for all active projects, features, and AI agent activity. It includes Shadow Engineering, a companion mode for teams adopting AI-assisted development incrementally.

​
Layout

Mission Control uses a two-panel layout:
  • Sidebar (left): Resizable project tree showing all configured projects with their active features/workspaces
  • Main panel (right): Feature activity view, planning artifacts, and status reviews

​
Project Tree

Projects appear as collapsible folders. Each active feature shows: State labels are computed from multiple signals: tmux session presence, heartbeat freshness (10-minute threshold), review/test status from the central review-status.json, and agent state files.

​
Release / Publish Indicator

The project HOME tab shows a compact Release row for projects with a publish pipeline: the most recent release-workflow run (live spinner while in progress, then ✓/✕ with the tag deep-linked to the GitHub Actions run), the latest GitHub Release, and each published npm package’s latest dist-tag linked to the registry. A partial failure — the npm publish succeeded but a build matrix job failed, so the GitHub Release was skipped — is called out explicitly with links to the failing jobs, since that shape is otherwise invisible. Per-project resolution lives in projects.yaml under publish (npm_packages, release_workflow), with inference fallbacks: the project root package.json name (when not private) and .github/workflows/release.yml (when present in the checkout). Projects with no publish pipeline simply don’t show the row. The surface is read-only — cutting a release stays in the CLI (pan release stable).

​
Activity View

When you click a feature, the main panel shows all agent sessions unified into one scrollable view. Each session appears as a collapsible section with:
  • Type badge: PLANNING (blue), WORK (green), REVIEW (amber), TEST (indigo), MERGE (pink)
  • Model: Which Claude/LLM model the agent used (e.g., Opus 4.6, Sonnet 4.5)
  • Duration: How long the session ran
  • Status dot: Green pulse (running), gray (completed), red (failed)
  • Transcript: Full agent output with tail-anchored auto-scroll for running sessions

​
Section Isolation Mode

Click a section header to view it full-screen. Press Esc or use the browser back button to return. The isolated view includes model display and a prominent keyboard hint.

​
Decisions

Agents stop and wait for you: a question, a plan to approve, a permission request, a rate-limit dialog, or simply a finished turn. Decisions is the one place that lists all of them, whether they come from a pipeline agent, a conversation, or the flywheel. Anything waiting on you is marked wherever it appears. A node awaiting your reply carries a pulsing amber indicator — click it to bring the dialog back, including one you dismissed with Esc — and the whole issue row is shaded so you can spot it without opening the tree. The list separates Blocking work from Waiting. Blocking means the agent has stopped until you answer, so it is burning wall-clock and nothing else will move; waiting means work continues around your decision. Within each group the oldest is first, since that is the one costing the most. Each row shows the actual question rather than just its kind, so you can often decide without opening anything. Answer re-opens the dialog for that subject. Dismissing a dialog never removes its entry here — that is the point of the surface. A dismissed question is still an open one.

​
Badge Bar

Quick-access badges appear below the feature header:

​
Status Reviews

Click the Status badge to generate an AI-powered progress review. The review:
  1. Gathers context: PRD, STATE.md, git diff, commit log, file changes, review/test status
  2. Sends everything to the configured LLM (set via Settings > Workflow Agents > Planning)
  3. Returns a structured analysis: summary, PRD coverage table, risk assessment, and recommendations
If no API key is configured, a static template with raw data is shown as fallback. Any configured provider (Anthropic, OpenAI, Google, Kimi, Z.AI) can be used for status reviews.

​
Shadow Engineering

​
Concept

Shadow Engineering is for teams that want AI assistance without replacing their existing workflow. Instead of AI doing the work, AI observes, documents, and assists.

​
How It Works

  1. Enable Shadow Mode: When creating a feature workspace, toggle “Shadow Engineering” on
  2. Monitoring Agent: Analyzes available artifacts (issue description, tracker comments, meeting transcripts, PR descriptions) and produces an Inference Document — the AI’s working understanding of what the team is building
  3. Inference Document: Not a prescriptive PRD but an inferred understanding. Surfaces gaps, ambiguities, and risks. Updates as new artifacts arrive
  4. Observer Agent: Watches the team’s actual development. Comments on PRs with observations and suggestions. Only commits code when explicitly asked

​
Value Proposition

AI that learns your team before it leads.
Shadow Engineering lets existing engineers keep working their way while Overdeck’s AI observes, documents, and assists. It learns the codebase, patterns, and team approach — so when you’re ready to go further, the AI already understands how you build. Ideal for:
  • Teams wanting to adopt AI-assisted development gradually
  • Organizations that need to maintain existing workflows during transition
  • Leaders who want AI oversight and documentation without AI-driven changes
  • Teams evaluating AI coding tools who need a low-risk entry point

​
Shadow Workspaces in the UI

Shadow workspaces are visually distinct in Mission Control:
  • Marked with a shadow indicator in the project tree
  • Show the Inference badge instead of generating PRDs
  • Inference modal includes an explanation of what the document represents

​
Planning Artifacts

Each feature workspace has a .planning/ directory:

​
Uploading Artifacts

Use the Upload badge to attach markdown or text files. Files are categorized as transcripts or notes and stored in the appropriate subdirectory.

​
Syncing Discussions

Click Sync to pull the latest comments from GitHub Issues or Linear. Discussions are converted to markdown and stored in .planning/discussions/.

​
Pipeline View

Where the project tree organizes work by project and feature, the Pipeline page reorganizes the same issues by lifecycle phase. A metric strip — Active issues, Work running, Review queue, Ship, and Spend — sits above phase-grouped rows (Ship, Review, Verifying), with a live activity feed on the right.
Use the project tree when you’re following a single feature’s agent conversation end to end, and the Pipeline view when you want to see every issue’s phase across all projects at once.

​
Configuration

​
Adding Projects

Projects are configured in ~/.overdeck/config.yaml. Any project with active workspaces appears in Mission Control’s project tree.

​
Model Selection

The model used for status reviews is configured in Settings > Workflow Agents > Planning. See Configuration for details.