Background agent results should be viewable like foreground agent output

Description

When subagents run in the foreground, their output (tool calls, intermediate results, final text) is visible in the conversation and can be expanded with ctrl+o. When subagents run in the background (run_in_background: true), only the completion notification is surfaced — the full agent output is invisible to the user.

The parent agent sees the result blob and can relay it, but the user has no way to directly review:

• What web searches the agent performed
• What files it read
• What intermediate reasoning it produced
• The full unedited final output

Current workaround

The raw JSONL conversation logs exist on disk at .claude/projects/.../subagents/agent-{id}.jsonl, but these are machine-readable, not human-readable. Users must write custom extraction scripts to review agent work.

Proposed behavior

Background agent results should be reviewable in the UI after completion, with the same level of detail as foreground agents (tool calls, results, final output). This could be:

1. An expandable section in the completion notification
2. A /agent-results {id} command
3. Automatic inline expansion (like ctrl+o for foreground agents)

Why this matters

Background agents are essential for parallelizing research-heavy tasks (running 6 competitor analysis agents simultaneously, for example). But the trade-off of losing all visibility into agent work is steep. Users want to verify agent reasoning, check sources, and review the full output — not just trust the parent agent's summary.

Environment

• Claude Code 2.1.92
• Windows 11
• Model: claude-opus-4-6 (1M context)