fix(tui_gateway): stop replaying live-turn user text after redirect

[StanleyStetson]

What does this PR do?

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 interrupt redirect). That re-fire produced a second agent turn and a second active role=user row with the same content (#84417).

Root cause (contract layer): while turn P was live, a text-only self-copy of P could sit in session["queued_prompt"] / queued_prompts (e.g. mid-busy re-submit while redirect was not available; consecutive-text merge could also glue P\n\nQ). Successful redirect/steer of Q did not scrub that self-duplicate, so post-settle _drain_queued_prompt restarted P.

Approach:

1. Scrub text-only self-duplicates of inflight_turn.user on successful redirect/steer; rewrite merged {P}\n\n{Q} → Q only.
2. Refuse admitting a text-only self-copy of the live user in _enqueue_prompt (scrub first so merge cannot re-glue P+Q).
3. On compression session rotation, bump _queued_prompt_generation so an in-flight drain claim cannot dispatch with a stale generation. If generation cancels a claim mid-drain, the claimed envelope is restored to the queue (legitimate follow-ups are not dropped).

Scope / residual (honest limits of this PR):

• Fixes server self-duplicates of the live turn. The independent Desktop client composer queue ($queuedPromptsBySession) is unchanged; a user-queued follow-up can still drain by design.
• Image-bearing queue envelopes are left alone (chronology/ownership).
• Verification is production-path unit/integration on gateway helpers (class repro), not a full interactive long-session Desktop compression e2e.

Related Issue

Fixes #84417

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✅ Tests (adding or improving test coverage)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• tui_gateway/server.py — _sanitize_queued_entry_vs_inflight_user, _drop_queued_duplicates_of_inflight_user; scrub on successful busy redirect/steer; refuse self-dup in _enqueue_prompt; bump _queued_prompt_generation in _sync_session_key_after_compress on rotation; restore claimed queue envelope when generation cancels mid-drain.
• tui_gateway/methods_session.py — scrub on successful session.redirect and session.steer (Desktop production path).
• tests/test_tui_gateway_queue_on_busy.py — redirect scrub, hard-interrupt path, merge rewrite, drain E2E (no second P), enqueue refuse, compress generation bump / no-op, generation-cancel restores claimed head.
• tests/test_tui_gateway_server.py — session.redirect RPC scrub + build-window queue of Q scrubs stale P.

How to Test

1. Regression suite:

   scripts/run_tests.sh tests/test_tui_gateway_queue_on_busy.py -q --tb=short
   scripts/run_tests.sh tests/test_tui_gateway_server.py -k "session_redirect" -q --tb=short
   scripts/run_tests.sh tests/tui_gateway/test_protocol.py -k "sync_session_key_after_compress" -q --tb=short

2. Evidence (local runner): tests/test_tui_gateway_queue_on_busy.py — 25 passed; related session.redirect / compress re-anchor tests passed via scripts/run_tests.sh.
3. Key cases:
   • successful redirect drops queued self-dup of inflight user (busy-submit + Desktop session.redirect RPC)
   • hard-interrupt + queue of Q scrubs stale P ahead of Q
   • build-window session.redirect enqueue of Q scrubs P
   • after redirect(Q), _drain_queued_prompt does not start a turn with P
   • merged P\n\nQ rewrites to Q-only
   • compress rotation bumps generation; unrelated follow-up text is preserved
   • generation cancel mid-drain does not drop the claimed follow-up (restored to queue)
4. Class repro (pre/post): P running + queued_prompt=P + redirect(Q) → pre-fix drain re-fired P; post-fix queue scrubbed, drain no-op.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run scripts/run_tests.sh on the affected tests and they pass (project canonical runner)
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: Linux (WSL/Ubuntu-class), via scripts/run_tests.sh

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) — pure Python gateway session-dict logic; no OS-specific paths
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

[spfcraze]

This was generated by AI during triage.

Summary:
A queued multi-line follow-up whose first line equals the live prompt is truncated to its later lines when any later redirect scrubs the queue.

Problems:

• In tui_gateway/server.py, _sanitize_queued_entry_vs_inflight_user iterates for sep in ("\n\n", "\n") (line 7612), treating a single newline as a merge boundary as well. _enqueue_prompt's consecutive-text merge only ever joins with \n\n (line 7579, f"{prev}\n\n{text}"), so the single-newline branch has no merge artifact to catch: it only fires on genuine multi-line user text.
• A queued follow-up like write a poem\nabout the sea under a live prompt write a poem is rewritten to about the sea on the next successful redirect or steer (each calls _drop_queued_duplicates_of_inflight_user, which runs this sanitize on every queued envelope), dropping the first line the user wrote.

Solution:
Restrict the rewrite to the merge separator _enqueue_prompt actually produces, the double newline, so a legitimate multi-line follow-up is never truncated.

Evidence

no deterministic fact backs this claim — model belief, not executed or read evidence

---

Checked against 7c39f7d — the tip of fix/84417-prompt-replay-one-commit when this was written — and fa83af3, main at the same moment.

[teknium1]

Thank you for this fix! It was salvaged into #86593 (cherry-picked onto current main with your authorship preserved in the commit history) and is now merged. Closing this PR since the work has landed.