Agent Viewer

A VS Code sidebar that shows every Claude Code agent running on your machine, grouped by Running / Idle / Done, with live status and one-click controls.

Claude Code writes a JSONL transcript for each session under ~/.claude/projects/. This extension watches that directory and surfaces what's happening across every project, so you can see at a glance which agents are active, what they're currently doing, and stop or clean up a session without leaving the editor.

Features

• Live dashboard — all sessions across the machine, auto-refreshed every 5s
• High-level status per agent — e.g. Bash: npm test, Edit auth.ts, Thinking, Awaiting response
• Click to expand any card to see:
  • Working directory
  • Session timing (started / last activity)
  • Latest user prompt
  • Recent tool calls (last 5, newest first)
  • Recent files touched (Read/Edit/Write targets)
  • Subagents spawned
• Per-agent actions — view transcript, open project folder, stop (SIGTERM), delete transcript

Install

Download the latest .vsix from the Releases page, then install it:

code --install-extension agent-viewer-<version>.vsix

Or from inside VS Code: ⌘⇧P → Extensions: Install from VSIX… → select the downloaded file.

Click the Agents icon in the Activity Bar once it's installed.

Build from source

git clone https://github.com/dipeshnx/agent-viewer.git
cd agent-viewer
npm install
npm run build
npx @vscode/vsce package
code --install-extension agent-viewer-0.0.1.vsix

Develop

npm install
npm run watch   # rebuild on change

Open the folder in VS Code and press F5 to launch the Extension Development Host.

How it works

• Watches ~/.claude/projects/**/*.jsonl with chokidar
• Tails the last ~64 KB of each JSONL and parses the last ~20 events
• Derives state from file mtime: < 10s → running, > 24h or terminator event → done, else idle
• Stop uses ps + lsof to match the Claude process by cwd; POSIX only (macOS / Linux)

Privacy

This extension reads the JSONL transcripts Claude Code writes to ~/.claude/projects/ — these may contain your prompts, assistant replies, and tool output (file contents, shell output, etc.). Everything stays local: no network calls, no telemetry, nothing is sent anywhere. The Stop action invokes ps / lsof locally to find the Claude process matching an agent's working directory.

Limitations

• macOS / Linux only for the Stop action. Everything else works cross-platform.
• "Started ago" reflects the earliest timestamp in the tailed window, not true session start, for very long sessions.
• Project-directory decoding from the ~/.claude/projects/ folder name is lossy; the extension prefers the cwd field embedded in events when present.

License

MIT