Agent tool docstring references SendMessage for resuming subagents, but SendMessage is not available in this runtime

Summary

The Agent tool's documentation includes:

"To continue a previously spawned agent, use SendMessage with the agent's ID or name as the to field — that resumes it with full context. A new Agent call starts a fresh agent with no memory of prior runs, so the prompt must be self-contained."

This text reads as guidance for resuming a truncated/incomplete subagent run. However, SendMessage is not registered as a tool in this runtime:

ToolSearch query="select:SendMessage" → "No matching deferred tools found"

There is no available mechanism to actually continue a previously spawned agent. The only option is to Agent(...) again with a self-contained prompt — i.e., a fresh agent that re-discovers any prior state from disk.

Reproduction

1. Inside a Claude Code session, attempt to use ToolSearch with query: "select:SendMessage", max_results: 1.
2. Observe the response: No matching deferred tools found.
3. Read the Agent tool docstring — note the SendMessage reference.

Expected

Either:

• (a) SendMessage is added/exposed in this runtime so the documented capability works; or
• (b) The docstring is updated to remove the SendMessage reference and document the actual recovery flow (e.g., "if a subagent terminates incomplete, dispatch a new agent with a continuation prompt; rely on filesystem state for recovery").

Why this matters

Combined with the related issue about background agents being truncated at ~20–30 tool uses (filed separately), there is currently no in-runtime way to recover a partially-completed agent's work other than:

1. Inspect the filesystem for any partial outputs the agent wrote
2. Manually craft a new prompt that picks up from that state
3. Dispatch a fresh agent

The advertised SendMessage path would be much cleaner if it existed.

Environment

• Claude Code CLI
• Linux (WSL2)