feat: inject queued messages into next API call for mid-task guidance

[shdomi8599]

Related GitHub Issue

Closes: #11802

Roo Code Task Context (Optional)

N/A

Description

TL;DR: Queued messages now get injected into the very next API call instead of waiting for the task to fully complete.

After discussing the approach with @hannesrudolph in DMs, we agreed that the simplest and most natural UX is to enhance the existing message queue rather than adding a separate steering mechanism. When users type a message while the agent is working, they expect it to be noticed at the next step — not after the entire task finishes.

Implementation (25 lines in Task.ts):

• At the finalUserContent assembly point in recursivelyMakeClineRequests, drain the message queue
• Append any pending messages as a [USER_MID_TASK_MESSAGE] block
• The agent reads them immediately at the next API request and incorporates the feedback

What's NOT changed:

• No new UI elements, buttons, or message types
• No new dependencies
• Existing ask blocking-point queue handling is preserved — messages queued during an API call are still handled by the existing logic at the ask blocking point

Design trade-off: This is "reactive" injection (between API calls), not "preemptive" (mid-stream abort). Preemptive injection is noted as future work. The reactive approach is zero-risk to existing behavior and covers the most common use case.

Test Procedure

Automated:

• npx vitest run core/task/__tests__/mid-task-message-injection.spec.ts — 5 tests covering: empty queue, single message injection, multiple message ordering, queue drain completeness, and post-drain independence
• Existing ask-queued-message-drain.spec.ts continues to pass

Manual (F5 Extension Development Host):

1. Start a long task: "Write a Python script to find primes from 1-100 with detailed comments"
2. While the agent is streaming, queue a message: "Actually, use JavaScript instead"
3. Observe: the agent picks up the message at the next API call and switches to JavaScript

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

No UI changes.

Documentation Updates

• No documentation updates are required.

Additional Notes

The original issue proposed using an external @steer-agent/core library. After feedback from @hannesrudolph ("Why not make that the default?"), the approach was simplified to a minimal change to the existing queue behavior — no external dependencies, no new UI, just 25 lines that make the queue smarter.

Future work: Preemptive injection — abort the current stream when a queued message arrives and restart with the message injected, enabling truly real-time mid-stream course correction.

Get in Touch

Discord: shdomi_8599

Interactively review PR in Roo Code Cloud