feat: Alt+Enter queues follow-up messages without interrupting agent

[iRonin]

Repurposes Alt+Enter from 'insert newline' to 'queue follow-up message'.

Behaviour:

• Enter (agent running) → interrupts/steers current response (unchanged)
• Alt+Enter → queues message into _pending_input — sent as the next turn, no interruption
• Ctrl+J → inserts newline for multi-line input (unchanged, was already bound)

UI feedback:

• Status bar shows 📬 N when follow-ups are queued
• Placeholder while agent runs: Enter to interrupt · Alt+Enter to queue follow-up · 📬 N queued
• Placeholder while idle with queue: 📬 N follow-ups queued — Alt+Enter to add more

Implementation:

• _followup_queue list mirrors _pending_input for display (popped in process_loop as messages are consumed)
• Alt+Enter never interrupts regardless of busy_input_mode config

[britrik]

Code Review: Alt+Enter Queue Feature

Thanks for this useful feature! I've reviewed the diff and found a few issues to address:

Critical Bugs

1. Cancel detection fails for messages with images (line ~7750):

   • In , payload can be either or tuple
   • In , only the text is added to
   • In , the check fails when payload is a tuple because it compares the tuple against a string

2. Duplicate text causes incorrect cancellation (line ~7711):

   • If two different follow-up messages have identical text, recalling one will incorrectly cancel both since cancellation uses simple string matching

Potential Issues

3. **Memory leak in ** (line ~7711):

   • The set grows indefinitely as users recall messages
   • It only gets cleared when a matching item is dequeued, but if messages are consumed without matching (e.g., different payloads), the set leaks

4. Empty text queued with images (line ~6739):

   • When is empty but exists, the message is still queued with empty string
   • Consider requiring non-empty text, or document this is intentional

Minor Suggestions

5. Status bar emoji compatibility (line ~1582):

   • The 📬 emoji may not render in all terminals; consider a text fallback like or

6. Use a more robust identifier for cancellation:

   • Consider using an incrementing ID or UUID per queued message instead of text for cancellation tracking

Overall

The feature is well-implemented with good UX feedback (status bar, placeholder text). The above issues are edge cases but should be fixed before merge.

[iRonin]

@britrik thank you for review! I was not sure how busy you are but I started using Hermes yesterday and I asked it to add a couple of features from Claude Code and Pi that I missed and it cranked the first PR without even asking lol
we're cooked! so good! I love it! great job!
I also use it extensively for legal drafting besides coding.
Will pass your review or ask it to check it itself.

[iRonin]

Thanks @britrik — great review. Pushed a fix commit (f02bcb1):

#1 & #3 (duplicate text / set leak) — fixed with UUID tags. Alt+Enter now wraps each payload as {"_followup_tag": uuid, "payload": ...}. Cancellation stores and matches UUIDs, so two identical messages never cancel each other, and the set is strictly bounded (one UUID per queued message, discarded on match).

Phantom queue pops — also fixed. The display sync (_followup_queue.pop(0)) now only fires for tagged items, not for regular Enter messages. Previously every item dequeued from _pending_input would pop the display mirror, including non-followup messages.

#1 (image payload cancel) — was already correct. Both the queuing side and the process_loop side extracted text via payload[0] if isinstance(payload, tuple), so image payloads were handled correctly. The UUID approach makes this moot regardless.

#4 (empty text + images) — intentional. Allows queuing an image-only message (attach image, no text, Alt+Enter). Documented in the commit.

#5 (emoji fallback) — not changing. The 📬 emoji is used consistently with other indicators already in the codebase (📌 for stash, etc.).

[alt-glitch]

Likely duplicate of #5504 / #8492 — same Alt+Enter follow-up queue feature.

[alt-glitch]

Likely duplicate of #5504 / #8492 — same Alt+Enter follow-up queue feature.

[teknium1]

Thanks for the focused follow-up-queue UX work. The requested Alt+Enter behavior is not on current main, but this branch needs adaptation before it can be assessed for merge.

Problems

• The change reassigns Alt+Enter, but current CLI documentation still promises it as a newline binding in website/docs/user-guide/cli.md:102, :216, and :240, plus website/docs/getting-started/quickstart.md:243 and hermes_cli/tips.py:57. Main implements that contract at cli.py:13564-13573.
• tests/test_cli_followup_queue.py only checks initialized attribute shapes. It does not exercise Alt+Enter/Alt+Up, queued payload consumption, cancellation, duplicate text, or image payloads.
• GitHub reports this branch as conflicting, and current busy-input routing is now at cli.py:13502-13560 with consumption at cli.py:15157-15338; the existing busy_input_mode="queue" path at cli.py:13525-13529 must remain intact.

Suggested changes

• Reconcile the keybinding/documentation contract and add behavioral keybinding/queue tests while adapting the implementation to the current routing paths.

Automated hermes-sweeper review.

[teknium1]

This file does not yet cover the stated Alt+Enter behavior: it only asserts constructor state. Please add behavioral tests that invoke the key handler and verify enqueueing, FIFO consumption, Alt+Up cancellation, duplicate text, and image payload cases.