Read what a Claude Code subagent actually did: its brief, every step and its result

The problem

When Claude Code hands a job to a subagent, the main conversation mostly sees the job going out and a summary coming back. What the subagent read, which commands it ran and what it edited on the way stay in a separate transcript file, so checking its work means trusting the summary or digging through JSON Lines on disk.

• A summary can say "all tests pass" without showing which tests ran.
• A failed or stopped subagent's reason is easy to miss in the parent's transcript.
• With several subagents running at once, you can't see which one is on which step.

How Pairbar solves it

1. Find the chat's subagents
   A 👥 chip on a chat's row counts the subagents working now. Click it, or choose Show subagents from the chat's menu, to list them under the chat: running first, then failed and stopped, then done (folded behind Show N done).
2. Hover for a quick look
   Rest the pointer on a subagent for its status, model and time, the tokens it used and its share of the chat's, how many tool calls it made, and what it's doing or found, with Open transcript and Copy result.
3. Open the transcript reader
   Click the subagent. The reader opens beside the sidebar: at the result for a done subagent, following along for a running one. Drag its edges to resize it; it keeps the size.
4. Read the brief, steps and result
   The brief is folded to three lines (Show more). Each step is one line, such as "Bash · swift test …" or "Edit · ChatStatus.swift (+24 −3)"; click one, or select it with ↑↓ and press ↩, for its output or diff. The result sits at the bottom with Copy.
5. Search it
   ⌘F searches the brief, the steps, their output and the result; ↩ or the arrows move between matches. Esc closes the search, then the reader.

Shortcuts

What the reader shows

Commands show as their program and subcommand ("npm test …") and links as their site and path, never their query. Show in chat opens the parent chat in Claude (the app's links can't point at a single message). Each subagent's menu also has Reveal transcript in Finder.

Outcomes

• Done: a finished answer.
• Failed: an error, with the reason (such as an API error or an unknown agent type), including a call that failed before the subagent started.
• Stopped: one you interrupted, or one cut off when its chat stopped.
• Background subagents (Claude's "run in background") show as running until they report back.

A subagent that's waiting for you makes its parent chat show Needs you. With Subagent finished or failed on (Settings → Notifications, off to start with), you get a banner when one ends while you're in another app. From the focused chat's context card, "3 running" unfolds the subagents and opens the newest in the reader.

Without Pairbar

Without Pairbar, subagent transcripts are still on disk and readable:

• Open the file: Claude Code keeps each subagent's transcript as agent-<id>.jsonl under the session's folder in ~/.claude/projects/, until its clean-up period (30 days by default) removes it.
• Filter it with jq, for example to pull out tool calls and the final message.
• Read the parent chat in the Claude app or terminal, which shows the job going out and the subagent's answer coming back.
• Resume the session with claude --resume to scroll back through it in the terminal.