Agent View: Orchestrate a Fleet of Claude Code Agents Without Leaving Your Editor

Claude Code happily runs many sessions and subagents at once. What it does not give you is a way to see them all in one place, know which one is blocked waiting on you, and steer them without juggling terminals. Agent View for Claude Code is an open-source VS Code panel that does exactly that: a live fleet list, a React detail view of any transcript, one-click spawning of parallel agents in isolated git worktrees, and a spatial canvas to watch it all happen. It runs in VS Code and every VS Code-based IDE — Antigravity, Cursor, Windsurf, VSCodium.

Run three or four Claude Code agents in parallel and the editor gives you no shared view of them. You cannot tell at a glance which one is thinking, which one is idle, which one hit an error, and which one is sitting blocked on a question. You end up tab-hunting across terminals to find the one that needs a decision. Agent View collapses that into a single Agents panel: a live tree of every session discovered from ~/.claude/projects, each row showing status (idle / running / thinking / needs-you), model, and token cost, backed by a React detail view of the full transcript.

The core safety primitive is the git worktree. One-click New Agent creates a fresh worktree and branch so parallel agents never clobber each other's files — each works on its own isolated copy of the repo. For agents the extension spawns, it passes claude --session-id so every agent maps deterministically to its transcript file, which is what makes the live view reliable rather than best-effort.

Parallelism is only useful if you can structure it. Agent View gives three structured modes on top of plain spawning.

Status is driven by Claude Code hooks that stream real-time events to a local server, with session-transcript replay as the zero-config fallback. Every agent shows an ambient "now doing X" one-liner derived from tool events — "Editing auth.ts", "Running: npm test" — with no extra LLM calls. When an agent needs your input, finishes, or errors, a toast (and optional chime) pings you, so you can walk away from a long run and get called back. A parent delegating to subagents shows a distinct thinking state with a count of its working children.

Three AI helpers ship off by default and consent-gated before first run. Conflict Radar is local-only and flags files edited by more than one agent. Attention Router triages which agent needs you into a "Needs you" inbox. Merge Advisor ranks multiple agents' diffs and recommends which to merge. The race never merges or deletes anything automatically — Pick winner just opens the diff and copies the git merge command, leaving the decision with you.

Published on both Open VSX (the default registry for Antigravity, Cursor, Windsurf, and VSCodium) and the VS Code Marketplace. Search "Agent View for Claude Code" in the Extensions view, or install from a terminal. You need Claude Code on your PATH and a Claude subscription.

# Open VSX-based IDEs (Antigravity / Cursor / Windsurf / VSCodium)
antigravity --install-extension inthepond.agent-view-for-claude-code

# VS Code (Microsoft build)
code --install-extension inthepond.agent-view-for-claude-code

We run agents in parallel across client projects every day, and the bottleneck was never the model — it was orchestration. Knowing which of six agents to look at, keeping their edits from colliding, and comparing competing attempts on the same task are coordination problems an editor should solve for you. Agent View is that missing layer. It is an unofficial community project, not affiliated with or endorsed by Anthropic; "Claude" and "Claude Code" are trademarks of Anthropic, used only to describe interoperability.