| name | user-preferences |
| description | Documented user preferences from feedback. Follow these when building new features. Covers cost/token display, UI conventions, and design preferences. |
User Preferences for Flightdeck
Documented preferences from user feedback. Follow these when building new features.
Cost & Token Display
- No monetary cost estimates — show token counts only, never dollar amounts.
- Token counts should be labeled as estimates:
~ prefix, (est.) suffix.
- Budget:
null or 0 means unlimited — don't show warnings for these values.
Alerts & Notifications
- No idle agent alerts — the Lead assigns tasks, not the user. Idle agents are normal.
- No context pressure alerts — Copilot handles compaction automatically.
- No predictions feature — agent stall prediction, cost prediction, context prediction are all handled by the system. Don't surface them.
Metrics & Analytics
- Remove metrics that don't account for task complexity:
- No model effectiveness charts (can't fairly compare when tasks vary in size).
- No role contribution charts (not a meaningful metric).
- No session score stars (subjective, not useful).
- Session history table is useful — keep it, but without stars.
Data & Historical Sessions
- Historical data on ALL pages — never show an empty state for existing projects. Every page must fall back to REST API data when live WebSocket data isn't available.
- Data retention/cleanup controls belong in Settings page.
Navigation & UX
- Everything that looks clickable must BE clickable — no decorative buttons or fake links.
- Unified project selection via tabs across all pages — use
<ProjectTabs>, not dropdowns.
- Milestones = progress events only — show task completions, progress reports, decisions, commits. Filter out agent spawn/termination/delegation noise.
Timeline
- Vertical scroll separate from horizontal — mouse wheel scrolls vertically only. Shift+wheel or trackpad horizontal gesture for horizontal pan.
- Default replay speed: 4× (not 1×).
- Replay scrubber must always be visible (sticky bottom, not clipped by overflow).
Chat & Communication
- User messages highest priority — always visible, never buried.
- Virtual scrolling for performance with large message histories.
- Group chats persist per project — don't lose them on session change.
General Philosophy
- Prefer showing real data with estimation disclaimers over showing "not available".
- Curate what's shown — less noise is better than more data.
- Consistent UX patterns everywhere — if one page uses tabs, all pages use tabs.
Typography
- JetBrains Mono for all monospace text, with ligatures enabled (
liga + calt).
- Font stack:
'JetBrains Mono', ui-monospace, monospace — ui-monospace handles CJK fallback natively.
- Always bundle fonts locally — no CDN dependencies. Must work offline.
Removed / Unwanted Features
- No predictions panel (agent stall/cost/context handled automatically).
- No model effectiveness chart (can't compare models across varying task sizes).
- No role contribution chart (not a meaningful metric).
- No session score stars (subjective, not useful).
- No monetary cost estimates — token counts only.
Release Process
- NEVER touch git tags — no create, no delete, no push. Tags are 100% user-managed. Only prepare version bumps in package.json and CHANGELOG updates. This caused an issue when v0.3.0 tag was pushed by an agent — user had to bump to 0.3.1. HARD RULE, no exceptions.
