Coherent plan mode: persist plans reliably and keep user-facing agent responsive when orchestrating background agents

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

Plan mode in the Claude Code CLI has two pain points that make it hard to use in real workflows.

1. Saving the plan to a markdown file is unreliable. I often start a planning session in a separate CLI session to prepare new work, intending to capture the plan as a .md file I can hand off or resume later. For some reason this saving step consistently becomes a bottleneck — the plan either doesn't get written cleanly, gets written in unexpected places, or the flow around finalizing and persisting the plan feels broken. As a result I've stopped using plan mode in the CLI, but then I'm exposed to the opposite problem: without plan mode, the agent tends to slip into code/execution mode when I just wanted to think.

1. Plan mode breaks the flow when one agent is orchestrating background agents. While the user-facing agent is in plan mode, it can't react to messages coming back from the background agents. Small decisions that the background agents need answered get deferred, the planning session stalls, and the parallelism that background agents are supposed to provide is wasted.

Proposed Solution

A "coherent plan mode" that:

• Reliably persists the resulting plan to a markdown file (configurable path, sensible default, clear confirmation of where it was written) so plans created in a dedicated planning session can be picked up by another session or another agent.
• Distinguishes "thinking/planning" from "executing code" without forcing the agent to be unresponsive. Specifically, when the user-facing agent is orchestrating background agents, plan mode should still allow it to receive and respond to messages from those background agents (at least small, non-destructive decisions) so the planning session doesn't block the workers.
• Optionally a "resume from plan.md" entry point so a plan written in one session can be loaded into another session and executed/refined without copy-pasting.

Alternative Solutions

Workarounds I've tried:

• Skipping plan mode entirely and just prompting carefully — risks the agent jumping into code mode when I only wanted to plan.
• Manually asking the agent to write the plan to a specific .md path at the end of a planning session — works inconsistently and adds friction every time.
• Running planning in a separate CLI session from execution — which is exactly the case where the broken plan-saving step hurts the most, because the whole point is to hand the plan off.

When orchestrating background agents there isn't a good workaround: plan mode currently makes the user-facing agent unable to answer the background agents' questions, so I either leave plan mode (and lose its benefits) or accept that the background agents stall.

Priority

High - Significant impact on productivity

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_