busymirror/ROADMAP.md

BusyMirror Roadmap

Shipped (highlights)

• Route-driven mirroring (multi-source)
• Accepted-only filter (mirror your accepted meetings)
• Persistent settings with autosave/restore; Import/Export JSON
• Overlap modes (allow, skipCovered, fillGaps) and merge-gap
• Work Hours filter and title-based skip filters
• Privacy: placeholders with prefix + customizable title
• 1.3.0: Mark Private option (global + per-route)
• 1.3.4: persistent file logging, stale-calendar pruning on refresh, clickable top-bar mode toggle
• 1.3.6: in-app scheduling via launchd with hourly/daily/weekday modes
• 1.3.6: generated macOS app icon set and packaged release assets
• 1.4.0: unit-test suite (45 tests), Cancel button, progress indicator, sandbox LaunchAgent fix, mirror URL fix, engine refactor into MirrorConfig
• Calendar list already auto-refreshes on EKEventStoreChanged (does not yet trigger an auto-sync — see Next)
• CLI diagnostics: --help, --list-calendars [--json], --status [--json], real exit codes (2 = no access, 3 = no saved routes), last-run tracking (time/ok/summary)

Next — reliability (V2 groundwork)

1. Event-driven background sync. The existing hourly launchd StartInterval only fires while the Mac happens to be awake at that instant and gets throttled/coalesced by macOS, so it's not reliable. Replace polling with reacting:
   • Keep the app running as a login item via SMAppService.agent instead of relying on launchd to relaunch it.
   • Extend the existing EKEventStoreChanged observer (today it only reloads the calendar list) to also trigger a debounced (~2-5s) auto-sync of saved routes.
   • Add an NSWorkspace.didWakeNotification handler to resync after sleep.
   • Keep one coarse fallback timer (e.g. every 30 min) purely as a safety net for a missed notification — not the primary mechanism.
   • launchd's remaining job shrinks to "make sure the app is running," which SMAppService likely covers, so the plist-generation code may become unnecessary.
2. MCP server (thin external wrapper, not embedded in the app). So agents driving BusyMirror don't have to shell out to the CLI and regex-parse log lines. A small standalone stdio-transport script (Node/Python) maps MCP tools 1:1 onto the CLI's --json output: list_calendars, list_routes, run_route, run_saved_routes, get_status. Deliberately kept out of the Swift app itself — no MCP SDK dependency in the signed binary (AGENTS.md's zero-external-packages rule stays intact), and MCP hosts spawn server processes on demand anyway, so there's no need for the app to run one persistently.

Then — V2 UI polish

• Split ContentView.swift (~1800 lines doing UI + settings + CLI + scheduling) into per-section view models (Routes, Schedule, Privacy, Log) so views are small and native-feeling.
• Real Settings { } scene instead of the main window doubling as preferences.
• Menu bar icon reflects state (idle / syncing / error) instead of a static icon.
• Surface last-sync-time / last-error / next-check in the menu bar UI — the data now exists (--status's lastRunAtISO/lastRunOK/lastRunSummary), this is just wiring it into the menu.
• Better server-side privacy mapping (per-provider heuristics).

Later

• Signed/notarized binaries and release pipeline.
• Smarter cleanup & conflict resolution.
• iOS/iPadOS helper (Shortcuts integration).
• Profiles & MDM/Managed Config support.

Decided against

• Direct Google/CalDAV/Exchange API integration (OAuth flows, token storage, per-provider clients so BusyMirror can mirror a non-local calendar without it being added to macOS). EventKit already surfaces any calendar added via System Settings → Internet Accounts — macOS does the sync, auth, and refresh. Building a parallel integration would duplicate the OS for the narrow case of an account that can't or won't be added system-wide (e.g. MDM-restricted work accounts). Not worth the OAuth/Keychain/per-provider-quirk surface for that.