🛰️ Agent Mission Control

A live dashboard, history viewer, and analytics chat that provides insights into the activity of GitHub Copilot CLI and GitHub Copilot App agent sessions. Tool calls, hook callbacks, sub-agent dispatches, skills, edits, reads and more become live signals so you can spot stalls, follow active work, and catch actionable provider issues without scrolling back through terminal output.

Built with Tauri 2, Phaser 4, and TypeScript.

🔗 Website: https://danwahlin.github.io/agent-mission-control

What it does

Agent Mission Control reads the local session state Copilot CLI already writes under ~/.copilot/session-state/ and tracks:

The dashboard panels show the selected session, recent activity, selected sector, replay controls, and an animated mission map. Open the session or sector inspector to filter recent calls by MCP, hooks, skills, sub-agents, or failures, then switch to the turn story to see what happened turn-by-turn. The Attention Center stays quiet for ordinary failed tool or hook completions and only surfaces actionable operational items such as provider scan problems or schema drift.

The topbar History route summarizes observed activity across all scanned sessions with 24-hour and local-day 7-day charts, model mix, event mix, top tools, token totals, recent sessions, and sanitized failure history. The Daily Log tab adds a calendar-driven daily debrief with "What I Worked On", models, token hotspots, tools/MCP, and high-activity sessions. Its Daily Digest export is an expandable, standup-ready summary — highlights of what you worked on per repo, by-the-numbers totals, models, and tools — so you can read the generated text before clicking the copy icon.

Analytics Chat lets you ask questions like "Where are my token hotspots?", "Which tools failed the most?", "What's my MCP server usage?", "Review my Copilot skills", or "What prompt improvements should I make?". It answers from indexed local analytics and, for prompt/skill/agent/MCP questions, can call the bundled read-only Mission Control Insights MCP tools to inspect bounded local definitions and inventory.

The normal activity bridge sends summarized activity to the renderer. Prompts, raw tool arguments, command output, file paths, and diffs are excluded from the live payload. The inspector has an explicit local-only reveal action for retained raw tool details when you need to inspect a selected call. Everything runs locally on your machine.

Focus mode

Click the 👁 button in the top-right to hide the side panels and put the mission map front-and-center. Great for a second monitor.

Install

Builds for macOS, Windows, and Linux are produced from the latest GitHub Release. The app is not code-signed; see the release notes for platform-specific quarantine/SmartScreen unlock instructions.

Develop

npm install
npm start            # builds frontend + launches Tauri dev window

Useful commands:

npm run build:frontend   # tsc + copy HTML/Phaser/assets to dist/
npm run build            # frontend + cargo build
npm test                 # build:frontend + playwright
npx playwright test      # tests only (build:frontend must run first)
cd src-tauri && cargo check

The frontend mounts at dist/index.html and Playwright serves it via python3 -m http.server 4173 --directory dist. Phaser runs inside the Tauri webview but is fully testable in plain Chromium because the __missionControlFixture global lets tests inject deterministic activity data.

Architecture (brief)

• src-tauri/src/agent.rs: AgentProvider trait + CopilotProvider impl. Scans ~/.copilot/session-state/, normalizes events through the bundled provider schema, caches the merged activity summary, and watches the directory with notify = 8 so the UI updates within ~300 ms of changes. Returns only allowlisted fields.
• src-tauri/src/lib.rs: Tauri commands (get_agent_activity, legacy get_copilot_activity, analytics/daily-log/chat commands, raw-detail reveal, editor/URL openers, app version, hide/quit) plus tray and single-instance wiring. Uses tauri-plugin-window-state to persist window position across launches.
• src-tauri/src/analytics.rs + mcp/mission-control-insights.ts: local analytics indexing, History/Daily Log summaries, Analytics Chat grounding, MCP server inventory, and read-only prompt/skill/agent inspection tools.
• src/scenes/MissionControl.ts: the single Phaser scene. Renders the mission map, nine sectors, pulse/arrival effects, ops status, responsive layout, and the space sprite atlas in assets/space/.
• src/main.ts: minimal Phaser bootstrap. One scene, opaque background, resizes with the window.
• src/index.html + hud.ts: slim 32 px top bar, route controls, History/Daily Log, Analytics Chat, reset/panels/theme controls, dashboard panels, replay controls, inspector dialog, Attention Center, and schema-drift dialog.

Assets

Sprites are drawn from a curated combined atlas under assets/space/ (see atlas.png + atlas.json). Source spritesheets live under assets/space/source/ for reference.

Releasing

npm run release 0.2.0

scripts/release.ts bumps package.json, src-tauri/tauri.conf.json, and src-tauri/Cargo.toml, regenerates CHANGELOG.md via git-cliff, commits, tags v0.2.0, and pushes. The Build & Release workflow then builds installers for all three platforms and attaches them to the GitHub Release.

License

MIT LICENSE.