
claude-agentpane
A side pane for Claude Code: the subagents a session runs, what each is doing, its tokens, and its conversation a click away.
Install with your AI
Paste into Claude Code, Cursor, or any agent — it reads the repo and wires the tool into your project.
Install and set up claude-agentpane (claude-plugin project) into my current project. Found on https://claudeers.com/claude-agentpane Repo: https://github.com/xuanji86/claude-agentpane Homepage/docs: — Detected install method: claude-plugin → /plugin install claude-agentpane@xuanji86/claude-agentpane Category: plugins. Platforms: cli, api, desktop, mobile. Read the repo's README for exact setup and env vars, then install it and wire it into my project. Claudeers Health Verdict: active; community-verified: false. Confirm the source before running anything.
/plugin marketplace add xuanji86/claude-agentpane /plugin install claude-agentpane@xuanji86/claude-agentpane
git clone https://github.com/xuanji86/claude-agentpane
// compatibility
| Platforms | cli, api, desktop, mobile |
|---|---|
| Operating systems | — |
| AI compatibility | claude |
| License | MIT |
| Pricing | open-source |
| Language | TypeScript |
agentpane
See your subagents at work, beside the conversation.
Every subagent a Claude Code session runs, what it is doing right now, what it costs in tokens, and its whole conversation one click away.
English · 中文
Why
When Claude fans work out to subagents, the terminal shows one line per agent and a footer list. agentpane is a mod that keeps them in a side pane, drawn the way Claude Code draws its own rows; click one and Claude Code's main view shows its conversation.
Features
| Live list | ⏺ Type(description), the tool call it runs now, +N more tool uses, and ✶ Running… (1m 12s · Opus 5.5 · 412k in · 1.9k out), or ✻ Waiting… while it waits on its own background work; listed the moment it spawns |
| Nested agents | An agent's own subagents sit indented under it |
| Tokens, model and effort | Input (cached share shown), output, context size and requests per agent, from the API's own usage figures; the model and thinking effort each agent's requests actually use; max_tokens in red when a response was cut off |
| Other loops | Model loops no listed agent claims (a workflow's agents, compaction and memory forks), counted under the list |
| Timeline | The latest batch of agents on one time axis, a bar each: what ran side by side, and what took longest |
| Batch receipt | Once a batch ends: ✓ 4 agents in 31s · 1m 18s of agent time (2.5× in parallel) · 21 tool uses · 190k tokens |
| Status line | ✻ 2 of 4 agents running under the prompt while the pane is folded or closed |
| Follows the main view | The agent you open from Claude Code's footer list is marked ◂ main view |
| Conversations | Click an agent: the main view switches to its transcript, and ◂ Main in the pane's header goes back. Where the main view can't switch (Claude Code before 2.1.300), it opens in the pane: its brief, replies in Markdown, tool calls with their results, runs of calls folded into one line |
| Stop | ■ Stop on a running agent's row, then Confirm stop, ends it through Claude Code's TaskStop |
| Notices | A toast when an agent finishes, one for several at once |
| Opens and folds itself | Opens when an agent starts, folds 10 s after the last one ends, to a tab above the prompt |
| Theme colours | Claude Code's own theme keys: follows dark, light and colourblind themes |
| Light on redraws | The spinner, shimmer, clocks and timeline run on the terminal's own frame clock; the pane redraws only when something changes |
| Main screen too | Without fullscreen, a summary of up to eight rows above the prompt instead of a side pane |
Install
Needs Claude Code 2.1.287 or later, the release that brought mods (function hooks); tested on 2.1.300. The side
pane needs the fullscreen layout (/tui fullscreen); on the main screen agentpane is a summary above the prompt. Mods
are early access: their API may change between releases, and a release that breaks the pane gets a fix here.
/plugin marketplace add xuanji86/claude-agentpane
/plugin install agentpane@claude-agentpane
From a terminal instead
claude plugin marketplace add xuanji86/claude-agentpane
claude plugin install agentpane@claude-agentpane
Use
|
A conversation in the pane — before Claude Code 2.1.300 Where the main view can't switch, clicking an agent opens its conversation here, and the pane widens to read it. |
Folded — the The pane hides and a tab above the prompt keeps the counts. Click it to bring the pane back; a new agent brings it back by itself. |
/agentpaneopens or closes the pane.3 donein the header hides the finished agents (their running children stay in sight).- The pane opens by itself where Claude Code seats a pane nobody asked for: at least 144 columns (110 once you have
opened it yourself). Below that,
/agentpaneopens it. - The keys
bkjwork while the pane holds the keyboard: it takes it when you open a conversation (over an empty prompt),Escgives it back.
Configure
In /config, or under pluginConfigs["agentpane"].options in settings.json:
| Option | Default | Meaning |
|---|---|---|
autoOpen | true | open the pane when an agent starts |
foldAfter | 10 | seconds after the last agent finishes before the pane folds; 0 keeps it open |
motion | true | animate the spinner and shimmer; off, they stand still and the clocks still run |
toasts | true | a toast when agents finish |
keepFinished | 8 | finished agents the list keeps (1–30) |
statusLine | true | the running agents under the prompt while the pane is not on screen |
What it can reach
Its footprint, as the awesome-claude-code-mods scan reads it from claude plugin validate:
agentpane watches, and acts only when you click an agent or press Stop. Every hook passes its event on unchanged.
| It sees | Through |
|---|---|
| the session's agents and their status | $.agent.list(), agent.spawn |
| each agent's tool calls (name and a short argument) | tool.call |
| each agent's token usage and stop reason | turn.step |
| an agent's transcript, while you have it open | $.session.messages({ agentId }) |
| which agent the main view shows | the pane's view |
the main view, switched to an agent you click (◂ Main back) | $.ui.view |
What it draws outside its pane: the tab above the prompt, toasts, and one status line ($.ui.status).
It makes no network requests, runs no processes, reads and writes no files, stores nothing across sessions and calls
no model. Its actions are yours: switching the main view on your click, and Stop: Claude Code's own TaskStop, on your second press, naming the agent in a cleaned,
shortened line. Text from transcripts is stripped of control, bidi and zero-width characters and cut to length before it
is drawn. claude plugin validate . prints exactly what it hooks and calls.
What is inferred, not measured
- Start times are when the pane first saw the agent: at its spawn, or, for one already running when agentpane loaded, the next poll.
- Tokens count from when agentpane loaded; an agent that finished before then shows none.
- Other loops cannot tell a workflow's agent from a compaction fork: both are model loops no listed agent claims.
- Tool counts are the calls the pane saw, and show only for agents it watched.
- A batch is the agents that ran with no quiet minute between them; the timeline and the receipt cover the latest.
Agent time is the sum of their run times, so
2.5× in parallelis that sum over the batch's wall time.
Limits
- The main view switches only on your click, and only from Claude Code 2.1.300 (
$.ui.view). Before that, and in Claude Desktop until its own copy of Claude Code catches up, the conversation opens in the pane instead. - Workflow agents have no names to list: the engine does not hand them to mods. They count under other loops.
Troubleshooting
The pane doesn't appear. Check claude --version is 2.1.287 or later, run /reload-plugins, then /agentpane.
Below 144 columns Claude Code won't seat a pane nobody asked for; /agentpane opens it at any width. Look in the
transcript for a dim line starting agentpane:: it names a hook that failed or why the pane was refused. Please
open an issue with it.
b k j do nothing. The pane needs the keyboard: click in it, or open a conversation over an empty prompt.
Too much motion. Set motion to false in /config.
Develop
git clone https://github.com/xuanji86/claude-agentpane
cd claude-agentpane
claude plugin validate .
claude plugin test .
claude --plugin-dir . # or add the folder to CLAUDE_CODE_PLUGIN_DIRS
A session that loaded the mod from its folder reloads it each time you save a file. The tests draw the pane on every
surface (terminal, desktop, VS Code, mobile) at 40–120 columns. python3 assets/make_previews.py redraws the README
pictures. Changes are listed in CHANGELOG.md. Issues and pull requests are welcome.
License
MIT © Anji Xu
// faq
What is claude-agentpane?
A side pane for Claude Code: the subagents a session runs, what each is doing, its tokens, and its conversation a click away.. It is open-source on GitHub.
Is claude-agentpane free to use?
claude-agentpane is open-source under the MIT license, so it is free to use.
What category does claude-agentpane belong to?
claude-agentpane is listed under plugins in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/claude-agentpane)
// retro hit counter
[](https://claudeers.com/claude-agentpane)
// reviews
// guestbook
// related in Claude Plugins
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explainin…
"CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/
financial-services — a Claude ecosystem project on GitHub.