See the Claude Code subagents and Codex spawned threads running inside your chats

The problem

A Claude Code subagent is a separate helper with its own context window that the main session hands a job to, for example "search the codebase for every caller of parseInvoice". That keeps the main conversation focused, but the work becomes harder to see: the helpers live inside the parent's transcript, and when you have several chats going there's no overview of which ones are busy with subagents and which are idle.

• You only see a subagent's progress by opening the parent chat and scrolling its transcript.
• When a helper stops to ask for a permission, the parent can look busy while nothing is moving.
• Codex threads that spawn other threads show up as separate entries, with no obvious link back to the thread that started them.

How Pairbar solves it

1. Look for the 👥 chip
   Rows with subagents working show a 👥 chip with a count, for example 👥 3. The count is how many are working now, not how many ever ran.
2. Open the subagent list
   Click the chip, or choose ⋯ → More › Show subagents. The list opens under the chat.
3. Read what each one is doing
   Chips on top count what's running, failed, stopped and done, with the tokens they used. Each entry shows its job, its type, how long it has run or took, and what it's doing now or the first line of its answer; running ones come first, then failed and stopped, and done ones fold behind Show N done. Nested subagents sit under the one that started them.
4. Read a subagent's transcript
   Click one to open the transcript reader beside the sidebar: its brief, each step on one line (opened for its output or diff), the result with Copy, search with ⌘F, and live follow while it runs.
5. Jump into a Codex spawned thread
   Codex subagents are real threads, so clicking one opens that thread in Codex.
6. Answer the one that's waiting
   A subagent waiting for your answer makes its parent chat show Needs you: the orange ? mark, a place in the Needs you section at the top, and a notification if you've turned alerts on.

Where subagent information shows up

Current actions are shown the same way as for chats: a shell command appears only as its program and subcommand (git push, never its arguments).

Defining the subagents themselves

Seeing subagents run is one half; the other is the agent definitions Claude Code picks from. Pairbar's Agents tab lists every personal, project and plugin agent Claude Code would load, and can write new ones for you. See build reusable Claude Code agents.

Without Pairbar

Without Pairbar you can still follow subagents; it just happens one chat at a time:

• Read the parent transcript in the Claude app or terminal. Each Task call and its result appear inline in the conversation.
• Use Claude Code's /agents command to list, create and edit subagent definitions from inside a session.
• Browse the Codex app's own thread list to find spawned threads, then open them one by one.
• Keep one terminal pane per session in a multiplexer such as tmux, and glance across panes.