Features outside the timeline/turn scope, prioritized for developers who use assisted coding daily (Codex, Claude Code, Cursor).
- Impact on the assisted-coding loop — fix prompts, review diffs, keep context, extend tools
- Usage frequency — what developers touch almost every turn vs. occasional scenarios
- Parity with the VS Code Codex extension — gaps that block adoption
- Effort vs. value — incremental deliveries with visible impact
Note: Features like voice, remote SSH, and browser automation have high market buzz but rank last here because they are peripheral for developers already working with a local editor agent — they do not block the core loop of prompt → plan → diff → approval.
| Feature | Why | Notes |
|---|---|---|
| Plugins & advanced MCP | Extensibility is the main differentiator vs. CLI; devs connect DBs, APIs, docs | Install UI, status, logs |
| Memories / persistent instructions | Context across threads reduces repeated setup | thread/goal, developer instructions |
| Message editing (rollback) | Fix a prompt without restarting the whole thread | thread/rollback + resend |
| Side conversations | Explore hypotheses without losing the main thread | thread/fork with excludeTurns |
| Rich diff review | Approve/reject hunks is the bottleneck in the agent loop | Side panel exists; review flow missing |
| Feature | Why | Notes |
|---|---|---|
| Multi-agent orchestration | Power-user interest | Status UI; handoff already in timeline |
| Kanban / issue integration | Team workflows (Linear, Jira) | Via MCP or custom API |
| Multi-user collaboration | Low demand in Codex clients today | Protocol still immature |
| Skills marketplace | Discovery and sharing | Depends on stable skill format |
| Local telemetry / analytics | Enterprise audit requests | Opt-in, privacy-first |
| Feature | Why | Notes |
|---|---|---|
| Themes and advanced accessibility | Inclusion and personal preference | i18n and theme base exists |
| Full keyboard shortcuts | Desktop power users | Map VS Code where it makes sense |
| Export conversation / report | Documentation and compliance | Markdown/PDF |
| Windows / macOS | Platform expansion | Tauri already abstracts part of the stack |
.deb / Flatpak installers |
Linux distribution beyond AppImage | Packaging and updates |
| Feature | Why | Notes |
|---|---|---|
| Realtime / voice mode | Trend in generic assistants; less critical for coding | Depends on stable App Server support |
| Remote SSH | Useful for remote machines; not the primary local flow | Host UX, sync, permissions |
| Git worktrees | Parallel tasks without polluting a branch; power users | Project + diff panel integration |
| Browser / computer use | Web QA and automation; niche within coding | Sandbox, permissions, preview |
- Turn summary (worked / stopped / failed)
- Live timer and divider before the agent reply
- Persistent plans (
todo-list) after restart - Dedicated activity cards (web search, MCP, subagent, compaction, etc.)
- Model-changed divider between turns
- Fork from an older turn (
thread/fork+thread/rollback) - Ephemeral commands and in-turn activity aggregation
Release 1.1: Rich diff review + message editing (rollback) — closes the main assisted-coding loop.
Release 1.2: Advanced plugins & MCP + persistent memories — less context friction and more agent capability.
Release 1.3+: Side conversations, then P1/P2 items based on demand.
Review this document each release or when the Codex App Server protocol publishes relevant breaking changes.