[BUG] Task tool usage guidance contradicts resumable subagent functionality (v2.0.28+)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

The Task tool's usage notes describe agents as "stateless" with "single message" returns, but this contradicts the resume parameter functionality documented in official sources.

Claude underutilizes resumable subagent capabilities because usage guidance suggests agents are one-shot operations, resulting in spawning new agents instead of resuming existing ones for multi-step investigations.

Evidence:

1. Official Documentation (https://code.claude.com/docs/en/sub-agents#resumable-subagents) describes resumable subagents that "continue previous agent conversations across multiple invocations while retaining full context"
2. Release Notes (v2.0.28): "Subagents: claude can now choose to resume subagents"
3. Tool Schema: Task tool includes working resume parameter accepting agent_id

What Should Happen?

Task tool usage guidance in system instruction should reflect that agents CAN be stateful when using the resume parameter (v2.0.28+), while maintaining backward compatibility notes about stateless behavior when resume is NOT used.

Suggested addition to usage notes:

• Agents can be RESUMED using the resume parameter (v2.0.28+). Pass the agent_id from a previous invocation to continue work with preserved context.
• When resuming is NOT used, each invocation is stateless.

Error Messages/Logs

N/A - No error messages. This is a documentation/guidance issue, not a runtime error.

Steps to Reproduce

Step 1: Simple Direct Query
Ask Claude Code agent: "What instructions do you have in Task tool's Usage Notes regarding agent state?"

Step 2: Expected Response
Claude will cite this from its instructions:
"Each agent invocation is stateless. You will not be able to send additional messages to the agent, nor will the agent be able to communicate with you outside of its final report."

<img width="1462" height="301" alt="Image" src="https://github.com/user-attachments/assets/96f1db7e-118c-4281-b2ec-3f17dd68631c" />

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.0.45

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

Impact:

• High priority affecting agent efficiency
• Multi-step investigations require spawning multiple new agents instead of resuming
• Loss of context and decreased performance in iterative workflows

Related Issues:

• #7317 - Original feature request for resumable subagents (implemented in v2.0.28)
• #11712 - Resume functionality currently broken (user prompts not saved to transcripts)

Note: This issue is about system guidance not reflecting the resume capability that was added in v2.0.28, separate from the implementation bugs in #11712.

Environment Details:

• Platform: Linux/WSL2
• Subscription: Claude Max