fix(tui_gateway): stop live-turn replay after redirect and preserve steer-mode message bursts

[teknium1]

What does this PR do?

Salvages and completes the Desktop steer-and-queued-messages cluster into one green branch:

1. Cherry-pick of #84589 (credit: @StanleyStetson) — fix(tui_gateway): stop replaying live-turn user text after redirect. Applied cleanly onto current origin/main with original authorship preserved. Stops the tui_gateway server queue from re-firing the live turn's original user prompt P after a mid-turn correction Q (Desktop session.redirect / busy-input redirect), which produced a duplicate agent turn and duplicate role=user row (#84417):

• Scrub text-only self-duplicates of inflight_turn.user on successful redirect/steer; rewrite merged {P}\n\n{Q} → Q.
• Refuse admitting a text-only self-copy of the live user in _enqueue_prompt.
• Bump _queued_prompt_generation on compression session rotation; a generation-cancelled drain claim is restored to the queue so legitimate follow-ups are never dropped.

2. New fix for #86134 — fix(tui_gateway): keep steer-mode fall-through bursts queued instead of interrupting.

Root cause: in _handle_busy_submit, the escalation guard was if mode != "queue" — so steer mode fall-throughs (agent steer() rejected, raised, or a non-steerable multimodal payload) fired _interrupt_busy_session. AIAgent.interrupt() both kills the live turn and clears the pending steer buffer, so a burst of user messages sent while the agent was busy could be silently destroyed: earlier successfully-steered messages vanished from the buffer and the in-flight work was killed. Exactly the reported symptom: "messages become steer / OOB and older asks get dropped."

Fix: steer-mode fall-throughs now keep pure queue semantics — preserved FIFO in queued_prompt / queued_prompts, drained on turn end via the existing _drain_queued_prompt path. Only explicit interrupt mode still escalates to a hard interrupt. Message-role alternation invariants are untouched: accepted steers continue through the sanctioned OOB steer-marker mechanism (apply_pending_steer_to_tool_results), and no synthetic user messages are injected mid-loop.

Related Issues

• Fixes [Bug]: post-compression Desktop session replays an earlier user prompt as a fresh turn #84417 (via fix(tui_gateway): stop replaying live-turn user text after redirect #84589 cherry-pick, credit @StanleyStetson)
• Fixes Desktop: queued burst of user messages dropped when treated as steer #86134

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✅ Tests (adding or improving test coverage)

Changes Made

• tui_gateway/server.py — fix(tui_gateway): stop replaying live-turn user text after redirect #84589's scrub/refuse/generation logic; Desktop: queued burst of user messages dropped when treated as steer #86134: mode != "queue" → mode == "interrupt" escalation guard with contract comment.
• tui_gateway/methods_session.py — (fix(tui_gateway): stop replaying live-turn user text after redirect #84589) scrub on successful session.redirect / session.steer RPCs.
• tests/test_tui_gateway_queue_on_busy.py — fix(tui_gateway): stop replaying live-turn user text after redirect #84589's suite plus 5 new Desktop: queued burst of user messages dropped when treated as steer #86134 regressions:
  • rejected steer queues without interrupting
  • steer exception falls back to queue without interrupting
  • multimodal payload in steer mode queues without interrupting
  • mixed burst: accepted steers preserved distinct + in order, fall-through queued, no interrupt
  • fall-through burst of 3 texts all drain FIFO after turn end (none dropped)
• tests/test_tui_gateway_server.py — (fix(tui_gateway): stop replaying live-turn user text after redirect #84589) session.redirect RPC scrub coverage.

How to Test

scripts/run_tests.sh tests/test_tui_gateway_queue_on_busy.py tests/test_tui_gateway_server.py

Evidence (local, venv pytest):

• tests/test_tui_gateway_queue_on_busy.py — 30 passed (25 from fix(tui_gateway): stop replaying live-turn user text after redirect #84589 + 5 new)
• tests/test_tui_gateway_queue_on_busy.py + tests/test_tui_gateway_server.py — 587 passed
• tests/tui_gateway/ — 450 passed; tests/test_tui_gateway_ws.py — 6 passed
• Red→green proof: all 5 new Desktop: queued burst of user messages dropped when treated as steer #86134 tests fail against pre-fix server.py, pass post-fix.
• ruff check clean; Windows-footgun checker clean; attribution audit: all emails mapped.

Salvage sweep

Open-PR sweep for #86134 ('queued burst', 'steer drop', 'busy queue desktop', 'queue follow-up') found no PR fixing this specific steer-path burst drop: #63298 (FIFO boundaries, different scope/architecture), #77892 (gateway platform debounce), #60555 (leftover steer marker), #65205 (desktop TurnQueue feature) are adjacent but none remove the steer-mode hard-interrupt escalation that destroys the burst. Implemented directly, minimal one-guard change.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs (sweep documented above)
• My PR contains only changes related to this fix
• Tests pass locally (see evidence above)
• I've added tests for my changes
• I've tested on my platform: Linux

Documentation & Housekeeping

• Docs — N/A (behavior fix, docstrings updated in-line)
• cli-config.yaml.example — N/A (no config keys changed)
• Cross-platform — pure Python gateway session-dict logic
• Tool descriptions/schemas — N/A

Infographic

Nous Research

[github-actions]

૮ >ﻌ< ა ci review

_{ran on 6e4d62d — fix(tui_gateway): keep steer-mode fall-through bursts queued}

⚠️ Warnings

CI timings · View report · View job

Wall time 4m39s vs 3m27s (+34.8%). 15 job(s) slower, 9 faster,

• Python tests / Run tests slice 1/12: -17.0s
• Python tests / Run tests slice 4/12: -12.0s
• Python tests / Run tests slice 3/12: -12.0s
• Python tests / Run tests slice 2/12: -12.0s
• OSV scan / Scan lockfiles / osv-scan: +12.0s

---

OSV vulnerability scan · View job

5 known vulnerabilities found in pinned dependencies.

• CVE-2026-67213 in package-lock.json
• CVE-2026-67213 in website/package-lock.json
• CVE-2026-70608 in package-lock.json
• CVE-2026-56876 in package-lock.json
• CVE-2026-70606 in package-lock.json

How to fix:

Review the findings in the Security tab. Update the affected dependencies if a patched version is available.