[FEATURE] Show tool logs for nested sub-agents spawned within an orchestrator session

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When an orchestrator agent spawns specialist sub-agents (software-architect,
full-stack-developer, code-reviewer, etc.) as nested calls within its own session,
the CLI output only shows the top-level orchestrator entry. The tool calls made by
each specialist agent (Read, Write, Search, etc.) are hidden inside the parent session
and never appear in the terminal log.

This makes it impossible to observe what each specialist is actually doing,
which files it reads, and how many tool uses it consumes — visibility that is
critical for debugging, cost monitoring, and trust in multi-agent workflows.

Proposed Solution

When a sub-agent is spawned inside an orchestrator session, stream its tool calls
to the CLI output with indentation or a prefix to indicate nesting level. For example:

● orchestrator(Coordinate mission)
● software-architect(Design backend)
Read(docs/requirements.md)
Read(docs/api-contracts.md)
Write(docs/architecture.md)
Done (14 tool uses · 84.3k tokens · 7m 58s)
● full-stack-developer(Implement backend)
Read(docs/architecture.md)
Write(server/server.js)
Done (11 tool uses · 71.3k tokens · 2m 20s)
Done (orchestrator)

Alternative Solutions

Current workaround: force the orchestrator to always output a "Delegation List"
and have Claude Code spawn each specialist as a separate top-level task.
This restores full log visibility but requires extra prompt engineering and
prevents the runtime from using native nested spawn capabilities.

A flag like --verbose-subagents or an environment variable would also be acceptable
as an opt-in alternative.

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

1. User sends mission: "Build a real-time WebSocket backend"
2. Claude Code spawns orchestrator
3. Orchestrator spawns software-architect, full-stack-developer, code-reviewer,

qa-engineer as nested agents

1. Currently: user sees only one log entry for the orchestrator — no visibility

into what any specialist did, which files were read/written, or token usage per agent

1. Expected: each specialist's tool calls stream to the terminal in real time,

indented under the parent orchestrator entry

Additional Context

_No response_