fix(neo-tui): surface every Bug 3 silent-failure path missed by PR #14

[code-yeongyu]

Follow-up to #14. Oracle's second skeptical sweep over the merged commit 45dbb577 found four more silent-failure leaks that all violate the original Bug 3 contract ("if there's an error, say so; if there isn't, say so"). Each fix is locked by a TDD-red-then-green regression test on top of the existing 298-test suite.

Defects fixed

1. Inbound::Response { success: false } was silently dropped

apply_inbound matched Inbound::Response(_) => {} and threw away every backend command rejection. The wire envelope already carries success: bool + error: Option<String> (see packages/neo-tui/src/rpc/envelope.rs::Response), so a backend-reported "submit failed" or "cycle_model failed" left the TUI looking healthy. Now pushes a Role::Error chat bubble (with the original error string when present, command-name fallback otherwise) and flips the footer to Status::Error / command failed. Successful responses still stay silent.

Tests: app_inbound_failed_response_surfaces_to_chat_and_footer, app_inbound_failed_response_without_error_message_still_surfaces, app_inbound_successful_response_does_not_disturb_chat_or_footer.

2. Inbound::ParseError only logged to tracing::warn!

A ParseError means the backend sent something the TUI can't decode. The previous code logged it and moved on, and the regression test app_inbound_parse_error_does_not_disrupt_ui actively asserted that the UI stays untouched. Both behaviors flipped: the runtime now pushes a chat error with the parser message + truncated frame preview and the footer goes to protocol error. Test renamed to app_inbound_parse_error_surfaces_to_chat_and_footer with inverted assertions.

3. senpi --neo -- --theme opencode/dracula failed with No such file or directory

Two compounding bugs:

• The bundled-theme registry stores flat keys (dracula, nord, ...) but the README documents the namespaced form (opencode/dracula).
• main.rs::looks_like_theme_path previously used value.contains('/') as the path heuristic, so any namespaced id was routed to read_to_string.

Fixed both: load_by_id strips the opencode/ prefix on lookup, and the path heuristic now only triggers on explicit filesystem indicators (./, ../, ~/, absolute path, or an existing file). Test: opencode_themes_are_resolvable_by_both_flat_id_and_opencode_prefix.

4. Alt+T (neo.theme.picker) was a no-op

The chord was bound in the JSON keymap and advertised in the README + help overlay, but execute_action had no match arm for it. The dispatcher silently fell into the catch-all Consumed no-op and the overlay never opened. Added AppAction::OpenThemePicker plus the dispatch path that builds ThemePickerOverlay::new(&self.theme.name). Test: alt_t_dispatches_open_theme_picker_and_opens_overlay.

Verification

• Rust: 305 tests passing (was 298 post-feat(neo-tui): full pi-tui rewrite — fixes all 4 reported bugs (shift+enter, Korean wrap, silent failure, idle/answering) #14), 0 failing, 1 ignored.
• Lint/format: cargo fmt --package senpi-neo-tui -- --check clean, cargo clippy --package senpi-neo-tui --all-targets -- -D warnings clean.
• TS regressions: the two neo-tui suites in packages/coding-agent/test/suite/regressions/ still pass 9/9 (keymap parity + arg forwarding).
• Workspace: npm run check (biome + tsgo + smoke + web-ui) clean across all 897 files.

Files

packages/coding-agent/CHANGELOG.md       |   1 +
packages/neo-tui/changes.md              |  15 +++
packages/neo-tui/src/app/mod.rs          |  40 ++++++++-
packages/neo-tui/src/main.rs             |  19 +++-
packages/neo-tui/src/theme/registry.rs   |  12 +-
packages/neo-tui/tests/app_loop.rs       | 135 +++++++++++++++++++++++++++++--
packages/neo-tui/tests/theme.rs          |  14 +++

Refs #14.

---

Summary by cubic

Surfaces every remaining neo-tui Bug 3 silent-failure path, completes picker, command, and exit UX, and surfaces extension UI notifications. Adds clear feedback for successful model/thinking changes and for non-rendered toggles.

• Bug Fixes
  • Backend failures: Response { success: false } and ParseError now push chat errors and set the footer to error.
  • Assistant/agent failures: message_end.errorMessage, failing compaction_end, and auto_retry_end { success: false } push chat errors; success paths stay quiet.
  • Transport/TTY failures: backend spawn, outbound serialization, stdout/stderr reader, and terminal input I/O errors surface immediately; disconnections are detected instead of freezing.
  • Extensions: extension_ui_request now routes to chat. notify shows a system note or a chat error (with footer error) for notifyType: "error". select/confirm/input/editor post a "not yet wired" note with the title. setStatus/setWidget/setTitle/set_editor_text stay silent.
  • Pickers: Ctrl+L opens the model picker; Alt+T opens the theme picker; picking a theme applies it or shows an error; picking a model posts a visible system note.
  • Theme IDs: registry accepts both dracula and opencode/dracula; CLI only treats explicit filesystem indicators (./, ../, ~/, absolute, or an existing file) as paths.
  • Unimplemented: advertised but unwired actions push a "not yet wired in --neo" system note instead of doing nothing.
  • Context scoping: tui.select.* outside overlays and tui.input.tab outside autocomplete show notes; neo.slash.open from the palette opens the slash overlay.
  • Exit: app.exit always quits, even with a non-empty input buffer.
  • Model/thinking changes: successful cycle_model, set_model, and cycle_thinking_level responses update header/footer and push a chat note; missing data shows an explanatory note.
  • Toggles: app.thinking.toggle and app.tools.expand now show a "not yet wired to chat rendering" note so the chord visibly lands.
  • Backward model cycling: app.model.cycleBackward reports "not yet wired" instead of cycling forward.

^{Written for commit f46f986. Summary will update on new commits. Review in cubic}

[cubic-dev-ai]

1 issue found across 7 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/neo-tui/src/main.rs">

<violation number="1" location="packages/neo-tui/src/main.rs:71">
P2: Absolute path detection is Unix-specific; Windows absolute paths can be misclassified as theme IDs.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Absolute path detection is Unix-specific; Windows absolute paths can be misclassified as theme IDs.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/neo-tui/src/main.rs, line 71:

<comment>Absolute path detection is Unix-specific; Windows absolute paths can be misclassified as theme IDs.</comment>

<file context>
@@ -54,6 +54,23 @@ struct Cli {
+        || value.starts_with("~/")
+        || value.starts_with("./")
+        || value.starts_with("../")
+        || value.starts_with('/')
+}
+
</file context>

Suggested change

	|| value.starts_with('/')
	|| Path::new(value).is_absolute()

[cubic-dev-ai]

1 issue found across 2 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/neo-tui/src/main.rs">

<violation number="1" location="packages/neo-tui/src/main.rs:71">
P2: Absolute path detection is Unix-specific; Windows absolute paths can be misclassified as theme IDs.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

1 issue found across 4 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/neo-tui/changes.md">

<violation number="1" location="packages/neo-tui/changes.md:97">
P3: This round-8 changelog entry is duplicated, which creates redundant/confusing release notes for the same fixes.</violation>
</file>

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

P3: This round-8 changelog entry is duplicated, which creates redundant/confusing release notes for the same fixes.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/neo-tui/changes.md, line 97:

<comment>This round-8 changelog entry is duplicated, which creates redundant/confusing release notes for the same fixes.</comment>

<file context>
@@ -74,6 +74,38 @@ Regression: `tui_select_action_outside_overlay_visibly_notifies_user`.
+
+Total Rust test count is now 321 passing (was 319 after round 7). `cargo fmt --check`, `cargo clippy --all-targets -- -D warnings`, and `npm run check` (897 files) all green.
+
+## 2026-05-19 — Oracle round 8: `neo.slash.open` + `tui.input.tab` outside their contexts
+
+Oracle's eighth review of the round-7 commit (`8ac34409`) confirmed the `tui.select.*` fix and all three gates, then flagged the next-most-similar leak: `neo.slash.open` was only handled in the raw key path (gated on Input focus + empty buffer for the mid-prompt `/` literal-insert behavior), so dispatching it through `execute_action` (i.e. via the command palette) fell into the catch-all silent consume. A follow-up classification audit also surfaced `tui.input.tab`, which `try_autocomplete_action` only consumes when an autocomplete popup is open — outside that context it landed in the catch-all too.
</file context>

[cubic-dev-ai]

2 issues found across 4 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/neo-tui/changes.md">

<violation number="1" location="packages/neo-tui/changes.md:157">
P3: The test accounting is inconsistent: this line says seven new tests but enumerates eight (and one is also listed as an existing updated test).</violation>
</file>

<file name="packages/neo-tui/src/app/mod.rs">

<violation number="1" location="packages/neo-tui/src/app/mod.rs:1091">
P2: `cycle_model` success handling updates model text but drops `thinkingLevel`, so thinking indicators can become stale after model changes.</violation>
</file>

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: cycle_model success handling updates model text but drops thinkingLevel, so thinking indicators can become stale after model changes.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/neo-tui/src/app/mod.rs, line 1091:

<comment>`cycle_model` success handling updates model text but drops `thinkingLevel`, so thinking indicators can become stale after model changes.</comment>

<file context>
@@ -1004,6 +1026,92 @@ impl App {
+        let provider = model_obj.get("provider").and_then(serde_json::Value::as_str);
+        let display = provider.map_or_else(|| name.to_owned(), |provider| format!("{provider}/{name}"));
+        self.header.model.clone_from(&display);
+        self.footer.model.clone_from(&display);
+        self.chat.push_system(format!("Model: {display}"));
+    }
</file context>

Suggested change

	self.footer.model.clone_from(&display);
	self.chat.push_system(format!("Model: {display}"));
	self.footer.model.clone_from(&display);
	if let Some(level) = data.get("thinkingLevel").and_then(serde_json::Value::as_str) {
	self.header.thinking_level = Some(level.to_owned());
	self.footer.thinking = Some(level.to_owned());
	}
	self.chat.push_system(format!("Model: {display}"));

[cubic-dev-ai]

P3: The test accounting is inconsistent: this line says seven new tests but enumerates eight (and one is also listed as an existing updated test).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/neo-tui/changes.md, line 157:

<comment>The test accounting is inconsistent: this line says seven new tests but enumerates eight (and one is also listed as an existing updated test).</comment>

<file context>
@@ -143,3 +143,15 @@ Refactor: extracted `apply_message_update_delta`, `apply_compaction_failure`, `a
+
+Refactor: simplified `AppAction::CycleModel { forward: bool }` → `AppAction::CycleModel` since the discriminator was always discarded by `action_to_command` and only `forward: true` is now reachable. Backward cycling never enters the variant; it lands at `note_unimplemented_action` in `execute_action`.
+
+Total Rust test count is now 329 passing (was 322 after round 9); seven new tests cover the three defects (`ctrl_t_app_thinking_toggle_visibly_notifies_user`, `ctrl_o_app_tools_expand_visibly_notifies_user`, `shift_ctrl_p_app_model_cycle_backward_visibly_notifies_user`, `apply_inbound_cycle_model_success_response_updates_displays`, `apply_inbound_cycle_model_success_response_with_null_data_pushes_note`, `apply_inbound_set_model_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_with_null_data_pushes_note`). Four existing tests updated to assert the new behavior (`ctrl_p_dispatches_cycle_model`, `shift_ctrl_p_app_model_cycle_backward_visibly_notifies_user`, `shift_ctrl_p_does_not_open_palette_after_rebind`, `ctrl_shift_p_no_longer_opens_palette_after_rebind`). One existing test replaced (`action_to_command_maps_cycle_model_regardless_of_direction` → `action_to_command_maps_cycle_model_to_cycle_model_command`). `cargo fmt --check`, `cargo clippy --package senpi-neo-tui --all-targets -- -D warnings`, and `npm run check` (897 files) all green.
</file context>

Suggested change

Total Rust test count is now 329 passing (was 322 after round 9); seven new tests cover the three defects (`ctrl_t_app_thinking_toggle_visibly_notifies_user`, `ctrl_o_app_tools_expand_visibly_notifies_user`, `shift_ctrl_p_app_model_cycle_backward_visibly_notifies_user`, `apply_inbound_cycle_model_success_response_updates_displays`, `apply_inbound_cycle_model_success_response_with_null_data_pushes_note`, `apply_inbound_set_model_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_with_null_data_pushes_note`). Four existing tests updated to assert the new behavior (`ctrl_p_dispatches_cycle_model`, `shift_ctrl_p_app_model_cycle_backward_visibly_notifies_user`, `shift_ctrl_p_does_not_open_palette_after_rebind`, `ctrl_shift_p_no_longer_opens_palette_after_rebind`). One existing test replaced (`action_to_command_maps_cycle_model_regardless_of_direction` → `action_to_command_maps_cycle_model_to_cycle_model_command`). `cargo fmt --check`, `cargo clippy --package senpi-neo-tui --all-targets -- -D warnings`, and `npm run check` (897 files) all green.

Total Rust test count is now 329 passing (was 322 after round 9); seven new tests cover the three defects (`ctrl_t_app_thinking_toggle_visibly_notifies_user`, `ctrl_o_app_tools_expand_visibly_notifies_user`, `apply_inbound_cycle_model_success_response_updates_displays`, `apply_inbound_cycle_model_success_response_with_null_data_pushes_note`, `apply_inbound_set_model_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_updates_displays`, `apply_inbound_cycle_thinking_level_success_response_with_null_data_pushes_note`). Four existing tests updated to assert the new behavior (`ctrl_p_dispatches_cycle_model`, `shift_ctrl_p_app_model_cycle_backward_visibly_notifies_user`, `shift_ctrl_p_does_not_open_palette_after_rebind`, `ctrl_shift_p_no_longer_opens_palette_after_rebind`). One existing test replaced (`action_to_command_maps_cycle_model_regardless_of_direction` → `action_to_command_maps_cycle_model_to_cycle_model_command`). `cargo fmt --check`, `cargo clippy --package senpi-neo-tui --all-targets -- -D warnings`, and `npm run check` (897 files) all green.

[code-yeongyu]

머지 완료. Oracle 12라운드 끝에 VERIFIED 받고 6efea12e로 squash-merge.

11라운드 동안 닫은 Bug 3 silent path 30+ 개 요약:

• R2: Inbound::Response { success: false } / Inbound::ParseError / opencode/ 테마 prefix / Alt+T 테마 픽커
• R3: Ctrl+L ModelPicker 오버레이 / 테마 픽커 selection apply
• R5: model picker selection note / 35-entry advertised-unimplemented list / spawn error visibility / writer + reader I/O
• R6: app.editor.external / app.suspend / message.errorMessage / CompactionEnd failure / AutoRetryEnd failure / EventStream::Err / stderr reader Err
• R7: tui.select.* outside an overlay
• R8: neo.slash.open from palette / tui.input.tab outside autocomplete
• R9: app.exit with non-empty buffer
• R10: app.thinking.toggle / app.tools.expand / cycle_model & cycle_thinking_level & set_model success-response data / app.model.cycleBackward silently cycled forward
• R11: extension_ui_request (notify / select / confirm / input / editor) frames

테스트 카운트: 298 (PR #14 머지 시점) → 333 passing, 0 failing, 1 ignored. cargo clippy --all-targets -- -D warnings clean, cargo fmt -- --check clean, npm run check (898 files) clean.

원래 4개 버그(shift+enter in tmux / 한글 폭 / 백엔드 silent failure / idle vs answering) 전부 lock 됐고, "에러가났으면 났다 안났으면 안났다 전혀안되노" 컨트랙트도 advertised 키맵 + RPC 전부 cover.

남은 follow-up (별도 이슈):

• /theme runtime swap UX
• 실제 GetAvailableModels consume + SetModel provider plumbing
• 세션 tree / fork / resume / compact
• 이미지 paste
• settings UI
• ~35개 advertised stub action (ADVERTISED_BUT_UNIMPLEMENTED_ACTIONS)
• Extension UI dialog 오버레이 (현재 timeout 처리에 의존)