AgentBar

Running coding agents in parallel has a characteristic failure mode, and it isn't the agent getting something wrong — it's the agent silently waiting. A session in a background terminal hits a permission prompt and blocks. You find out twenty minutes later, when you switch tabs and realize nothing has happened since the last time you looked. Multiply that by three or four concurrent sessions across different agents, and babysitting terminals becomes the actual job.

The existing answers don't scale: per-agent notifier apps (one more menu bar item each), notification spam you learn to ignore, or just alt-tabbing on a timer. What I wanted was one glanceable spot that always shows the session that needs me most — and, for the most common interruption of all, the permission prompt, the ability to answer right there without leaving whatever I was doing.

The whole system is files. Tiny Node.js hook scripts write one JSON file per session to ~/.agentbar/state.d/; the app watches that folder and renders. Claude Code reports with full fidelity (prompt, tool use, permission, stop, session lifecycle), Codex reports turn-complete via its notify hook. Copilot and Antigravity have no public event API yet — they're launchable from the menu and their mascots are ready for the day one appears.

Permission approvals are two more folders of the same protocol: the blocking PermissionRequest hook writes the request into requests.d/, the app writes the decision into answers.d/, the hook unblocks. When the bar shows more than it can say, priority decides: permission outranks working, working outranks idle. Every live session gets a menu row with project name, git branch and state — click it to jump to its app or terminal.

When a Claude Code session asks for permission, the request appears under the yellow needs approval row: exactly what's requested — e.g. Bash: git push origin main, full input in the tooltip — plus an inline button strip. ✓ Allow once, ✓ Always (only when Claude Code itself suggests a rule, shown as readable text like Bash(git push:*), never raw JSON), ✕ Deny, or hand the prompt back to the terminal or the Claude desktop app. Decisions return through Claude Code's own PermissionRequest hook, so the terminal prompt never even appears.

The design constraint that shaped all of it: the app must never become a hostage. If AgentBar isn't running, quits mid-wait, or you simply ignore the request for ten minutes, the prompt shows in the terminal exactly as it would have without AgentBar. Fail-open by construction — every failure mode was enumerated and each one ends at the ordinary terminal prompt. Codex and Copilot have no decision hooks, so their rows offer a clearly-labeled best effort instead: focus the session's terminal and press the approval key, gated behind macOS's Accessibility permission.

AgentBar is a clean-room rewrite of an earlier single-agent notifier concept, rebuilt around one bet: the simplest possible transport wins. State is plain JSON files in watched folders — the whole protocol is debuggable with ls and cat, survives crashes on either side, and adding an agent means adding a script that writes a file.

AgentBar is open source under MIT, at v1.4.0 — five releases shipped on day one, from the first multi-agent bar through remote Allow/Deny, the inline approval strip, a race fix and the question state. It installs with a one-line curl, a Homebrew cask, or by pasting the install command into your coding agent and letting it do the work. Hook mapping is documented as a spec — claude-code-states.md records which Claude Code events map to which states, including what's deliberately ignored and why.

Next levers: a notarized build (the app is ad-hoc signed today, so a manual download triggers Gatekeeper's warning), and live status for Copilot and Antigravity the moment either ships a public event API — the bar, the protocol and the mascots are already waiting.