fix(shell): remove queued-input bordered ghost in running-turn TUI

[elkaix]

Summary

Pressing Enter during an active agent turn to queue a follow-up left a redundant bordered "ghost" copy of the accepted input in terminal scrollback, beside the intended teal ❯ … queued row with its ↑ to edit · ctrl-s to send immediately hint.

Root cause: the running-turn lifecycle set Application.full_screen=True mid-turn — which prompt_toolkit 3.0.52 ignores for the renderer once the session is constructed, so it never achieved its purpose — while _sync_erase_when_done() gated final-frame erasure on not full_screen. Enter therefore final-rendered the accepted buffer as a fossilized bordered card.

Changes

• _sync_erase_when_done() now depends only on prompt mode: AGENT erases the accepted frame; SHELL keeps it in normal scrollback.
• Removed all dynamic prompt_toolkit fullscreen toggling (_set_running_fullscreen, _previous_full_screen) and their turn-start/attach/detach calls. _turn_starting, delegate attachment, invalidation order, and pre-stream input-card gating are unchanged.
• tui.sticky_input config key retained for backward compatibility with honest legacy wording; the composer stays inline during active turns.

Tests

• Rewrote the fullscreen-toggle unit tests into a screen-mode-preserved running-prompt lifecycle contract: AGENT-mode erasure stays enabled even when the application/renderer full_screen fields disagree, and the lifecycle never mutates those fields.
• Strengthened the Enter-to-queue test — exactly one queue append, one prompt invalidation, no steer/echo.
• Added a pyte-rendered PTY regression test_mid_turn_queued_input_renders_once_and_executes_once: the queued follow-up shows its live ❯ … / ↑ to edit row and never fossilizes as a bordered accepted-input card (the ghost), and executes once in FIFO order. The guard flags the queued text only when it is wedged between a card border directly above and committed ⏺ content directly below — the fossil signature — which excludes the two healthy renderings (the live queued display and the post-drain execute-echo). A/B-verified across every rendered frame: fails on the pre-fix code, passes on the fix.

Verification

• make check-pythinker-code (ruff + format + pyright 0 errors + ty) — green.
• make test-pythinker-code (unit + e2e including the local PTY suite) — green.
• docs/en/release-notes/changelog.md regenerated via npm run sync.

Summary by CodeRabbit

• Bug Fixes

  • Fixed a TUI issue where queued follow-up input could appear as a bordered ghost or duplicate row.
  • Queued input now displays once during active turns, executes once, and restores the normal input layout afterward.
  • Preserved prompt display state while attaching and detaching running prompts.

• Documentation

  • Updated configuration guidance and the changelog to reflect the corrected queued-input behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 9b54cfd2-6ac0-4efe-9447-a21513aee59f

📥 Commits

Reviewing files that changed from the base of the PR and between f2e03b1 and 6c7574d.

⛔ Files ignored due to path filters (1)

• docs/en/release-notes/changelog.md is excluded by !docs/**

📒 Files selected for processing (6)

• CHANGELOG.md
• src/pythinker_code/config.py
• src/pythinker_code/ui/shell/prompt.py
• tests/e2e/test_shell_pty_prompt_layout_e2e.py
• tests/ui_and_conv/test_prompt_tips.py
• tests/ui_and_conv/test_visualize_running_prompt.py

---

📝 Walkthrough

Walkthrough

Changes

Queued input lifecycle

Layer / File(s)	Summary
Preserve prompt screen state src/pythinker_code/ui/shell/prompt.py, src/pythinker_code/config.py, tests/ui_and_conv/*	CustomPromptSession no longer toggles fullscreen state during turn lifecycle transitions. Tests verify screen state preservation, erase behavior, invalidation, and default queued-input handling.
Validate queued follow-up rendering tests/e2e/test_shell_pty_prompt_layout_e2e.py, CHANGELOG.md	PTY coverage verifies queued follow-up input renders once, executes once, avoids bordered ghost rows, and restores the idle input-card border. The changelog records the fix.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Suggested labels: bug

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 40.00% which is insufficient. The required threshold is 70.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title follows conventional commit format and accurately summarizes the TUI queued-input ghost fix.
Description check	✅ Passed	The description covers the summary, changes, tests, and verification; only the issue link and checklist sections are omitted.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/queued-input-ghost

---

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!