Redesign Tau TUI as a transcript-first operator terminal

[njfio]

Problem statement

Tau TUI is still fundamentally an operator/debug shell and log renderer instead of a modern transcript-first terminal agent interface. It uses space poorly, exposes low-level runtime detail instead of task context, and does not yet deliver the interaction quality expected from tools like Claude Code, Codex CLI, or OpenCode.

Acceptance criteria

1. Interaction model defined

• Define transcript-first layout, compact status context, approvals, slash commands, interrupts, retries, tool/task visibility, and artifact/memory/cortex access.

2. Layout/system behavior defined

• Define resizing, focus, scroll, expansion, keyboard shortcuts, status line behavior, and command discoverability.

3. Progress and failure UX defined

• Define how in-flight model/tool/task activity is shown in-place, and how failures/recovery actions are rendered without collapsing into log spam.

4. Shared-state dependency honored

• TUI redesign must explicitly consume the shared operator turn/task state protocol rather than inventing a separate UI-only event model.

5. Test plan defined

• Define PTY/TUI regression and interaction tests for input, cancel, approvals, retries, tool streaming, and resize behavior.

Non-goals

• Webchat redesign.
• Swarm orchestration UX beyond the hooks the TUI needs.
• Replacing TUI with a browser-only workflow.

[njfio]

Spec branch pushed: 3582-runtime-timeout-cancel-recovery at 8efee510. Spec link: https://github.com/njfio/Tau/blob/3582-runtime-timeout-cancel-recovery/specs/3582-runtime-timeout-cancel-partial-output-recovery.md

[njfio]

Correcting the planning branch/spec mismatch for #3582.

Current branch: 3582-tui-transcript-first-redesign at 96ba3c53
Spec: https://github.com/njfio/Tau/blob/3582-tui-transcript-first-redesign/specs/3582-tui-transcript-first-operator-terminal.md
Research note: https://github.com/njfio/Tau/blob/3582-tui-transcript-first-redesign/docs/research/cli-interface-patterns-2026-03-16.md

This supersedes the earlier comment that pointed #3582 at the timeout-recovery branch/spec. The updated spec is now explicitly grounded in OpenCode interaction patterns: transcript-first layout, compact context bar, composer-led command discoverability, detail toggles, session/share actions, and interactive permission prompts.

[njfio]

First implementation slice landed on 3582-tui-transcript-first-redesign through cd5c6b06.

What changed:

• Corrected the issue/spec/branch mismatch and added OpenCode research anchors.
• Added red render tests for the first visible contract delta: transcript-first default, compact top context, and action-oriented composer hints.
• Shipped the first green layout slice:
  • detail drawer collapsed by default
  • status bar now includes session and approval context
  • composer/help hints now expose details, retry, and interrupt
  • transcript terminology replaces the old chat-shell posture
• Refactored prompt/system message updates to reduce duplication in the transcript path.

Validation:

• cargo test -p tau-tui

This is not the full OpenCode-style redesign yet. It is the first baseline shift away from panel-first shell behavior so the next slices can add the real transcript shell, right-side detail drawer, session/share affordances, and permission action strips on top of a cleaner default.

[njfio]

Implemented the next transcript-first TUI slice on 3582-tui-transcript-first-redesign and pushed the new commits.

New commits:

• 10269446 test(3582): red tests for transcript shell renderer
• 88b82252 feat(3582): add transcript shell renderer
• b439c8eb refactor(3582): split interactive tui modules
• 31ae4952 integrate(3582): cover interactive tui entrypoint smoke path

What landed:

• transcript-first body with a live activity strip above messages
• prompt-shell composer replacing the old bordered composer panel
• contextual details drawer with tools/memory/cortex/sessions sections
• compact top context now includes cwd/workspace alongside model/session/profile/approval/health
• renderer and interactive app loop split into focused modules under crates/tau-tui/src/interactive/
• integration smoke coverage for tau-tui interactive non-TTY failure path

Validation:

• cargo test -p tau-tui
• manual PTY smoke: cargo run -p tau-tui -- interactive --profile ops-interactive

Issue remains open. Remaining work for #3582 is the larger visual redesign: richer transcript cards, explicit approval controls, on-demand drawers beyond placeholders, and stronger parity with OpenCode interaction patterns.

[njfio]

Implemented the next #3582 slice on 3582-tui-transcript-first-redesign and pushed it.

New commits:

• bb82ef66 test(3582): red tests for attention strip and responsive drawer
• 9fb2f1f9 feat(3582): add responsive details drawer and attention strip
• 1a9e0f92 refactor(3582): split transcript activity and focus helpers
• dfe055ae integrate(3582): exercise drawer context via interactive input path

What landed:

• the details drawer now uses tabbed context navigation instead of a static list dump
• the transcript renders an inline failure attention strip with next actions
• the body now collapses the drawer first on narrow terminals to preserve transcript + composer space
• detail section state is explicit (tools, memory, cortex, sessions) and is reachable through real slash-command input (/memory, /cortex, /sessions, /tools)
• activity/attention rendering and focus helpers were split into smaller modules to stay within the refactor gate

Validation:

• cargo test -p tau-tui integration_spec_3582_memory_command_switches_detail_context_through_real_input_path
• cargo test -p tau-tui

Issue remains open. The next meaningful slice is still visual depth and approval UX: richer transcript cards, explicit approval controls, and stronger OpenCode-style spacing/hierarchy.

[njfio]

Implemented the next #3582 TUI slice on 3582-tui-transcript-first-redesign and pushed it.

New commits:

• f03bfaa0 test(3582): red tests for approvals cards and detail overlay
• a5295101 feat(3582): add approval strip overlay and transcript cards
• 1dc56040 refactor(3582): split interactive navigation helpers
• 75d0d6a9 integrate(3582): cover approval flow through interactive input path

What landed:

• explicit approval state and action strip with Approve / Reject actions
• slash-command driven approval flow: /approval-needed, /approve, /reject
• transcript messages now render as clearer card-style entries instead of raw line stacks
• narrow terminals now show a Quick details overlay instead of silently dropping detail access
• live smoke verified through the real tau-tui interactive PTY entrypoint

Validation:

• cargo test -p tau-tui red_spec_3582_
• cargo test -p tau-tui integration_spec_3582_approve_command_resolves_pending_approval_through_real_input_path
• cargo test -p tau-tui
• manual PTY smoke: cargo run -p tau-tui -- interactive --profile ops-interactive

Issue remains open. The next meaningful #3582 work is deeper polish and parity: better composer/shortcut hierarchy, richer detail surfaces than placeholders, and more deliberate visual spacing/contrast across the whole terminal shell.

[njfio]

Implemented the next #3582 slice and pushed it.

Commits:

• 432e3e61 test(3582): red tests for composer chips and detail summaries
• 385b902b feat(3582): improve tui composer approvals and detail summaries
• 618015d2 refactor(3582): simplify attention and detail rendering
• $(git rev-parse --short HEAD) integrate(3582): wire interactive shell composer and detail affordances

What landed:

• composer footer now uses action-chip style hints instead of a prose instruction sentence
• insert mode now resolves approval requests directly with y / n
• approval attention strip exposes both verb labels and shortcut labels
• memory details now surface degraded shared-state status explicitly
• sessions details now show live session metrics: messages, tokens, approvals pending

Validation:

• cargo test -p tau-tui
• manual TTY smoke via cargo run -p tau-tui -- interactive --profile ops-interactive
  • exercised /approval-needed, y, /sessions, Ctrl+C

Issue remains open because the larger transcript-first/OpenCode-quality redesign is still in progress.

[njfio]

Implemented the next #3582 slice and pushed it.

Commits:

• 9aa2820b test(3582): red tests for transcript run-state cards
• f1c2dce3 feat(3582): add transcript run-state cards
• 613ffa72 refactor(3582): split transcript tui tests by surface
• $(git rev-parse --short HEAD) integrate(3582): validate transcript run-state cards in interactive shell

What landed:

• transcript now shows a compact run-state card between activity and message history
• active turns surface prompt context and state labels inline
• running tool work surfaces tool name and detail without forcing the drawer open
• completed turns surface a last-turn summary card after prompt submission
• oversized ui_tests.rs was split into surface-specific modules to keep the test harness maintainable

Validation:

• cargo test -p tau-tui
• manual TTY smoke via cargo run -p tau-tui -- interactive --profile ops-interactive
  • verified prompt submission renders Last turn card inline
  • verified Ctrl+C exits cleanly

Issue remains open because the broader OpenCode-quality TUI redesign is still in progress.

[njfio]

Implemented the next #3582 slice and pushed it.

Commits:

• 116491d0 test(3582): red tests for live execution blocks
• 5157a95e feat(3582): enrich live execution blocks in transcript shell
• 4b40d266 refactor(3582): share transcript shell ui primitives
• $(git rev-parse --short HEAD) integrate(3582): validate live execution blocks in interactive shell

What landed:

• live activity strip now includes a state chip plus active tool counts and current tool name
• streaming turns now surface assistant-preview context inside the run-state card
• failure attention strip now exposes keyboard-first recovery hints ([r] retry, [/details])
• shared transcript-shell UI primitives were extracted so badge/action styling is not duplicated

Validation:

• cargo test -p tau-tui
• manual TTY smoke via cargo run -p tau-tui -- interactive --profile ops-interactive
  • submitted testing and verified inline run-state card behavior
  • exercised /approval-needed and verified approval attention state renders in the real shell
  • verified Ctrl+C exits cleanly

Issue remains open because the broader OpenCode-quality transcript shell and richer shared-state drawer content are still in progress.

[njfio]

Implemented the next #3582 slice and pushed it.

Commits:

• 35ac8e63 test(3582): red tests for drawer state summaries
• b41a0914 feat(3582): add real drawer state summaries
• 7cc42d53 refactor(3582): move transcript summary queries into chat panel
• $(git rev-parse --short HEAD) integrate(3582): validate drawer summaries in interactive shell

What landed:

• memory drawer now surfaces recent user context from the live transcript while still making degraded shared-state status explicit
• cortex drawer now surfaces observer/runtime posture from actual app state: current state, active tools, approval state, and current tool when present
• sessions drawer now exposes richer session summaries, including last prompt and assistant message count
• transcript summary queries were moved into ChatPanel so drawer rendering no longer digs through raw message vectors directly

Validation:

• cargo test -p tau-tui
• manual TTY smoke via cargo run -p tau-tui -- interactive --profile ops-interactive
  • submitted I like Python, Rust, and TypeScript.
  • verified /memory surfaces recent user context in the real shell
  • verified /sessions surfaces last prompt and assistant count
  • verified /approval-needed + /cortex surfaces approval/runtime posture
  • verified Ctrl+C exits cleanly

Issue remains open because the broader OpenCode-quality shell, denser transcript layout, and deeper shared-state consumption are still in progress.

[njfio]

Implemented the next TUI slice on .

What landed:

• typing from an empty composer now opens the command palette directly instead of inserting raw text
• added transcript actions through the real command surface:
  • exports the current transcript to
  • copies the current transcript through the configured clipboard command
• command palette defaults now surface transcript/share actions instead of hiding them behind a long command list
• transcript export/copy actions are wired through the same interactive input path as the rest of the shell
• clipboard/export behavior was split into focused modules to stay inside the refactor gate

Validation:

running 90 tests
test interactive::gateway_tests::red_spec_3582_gateway_response_events_drive_interactive_app_state ... ok
test interactive::gateway_tests::red_spec_3582_gateway_failed_events_drive_interactive_error_state ... ok
test interactive::gateway_tests::red_spec_3582_gateway_sse_parser_preserves_response_operator_state_contract ... ok
test interactive::ui::tests::approval::integration_spec_3582_approval_shortcuts_resolve_request_without_leaving_insert_mode ... ok
test interactive::ui::tests::approval::red_spec_3582_approval_attention_strip_exposes_approve_and_reject_actions ... ok
test interactive::ui::tests::approval::integration_spec_3582_approve_command_resolves_pending_approval_through_real_input_path ... ok
test interactive::ui::tests::composer::integration_spec_3582_bare_thinking_command_opens_overlay_through_real_input_path ... ok
test interactive::ui::tests::composer::integration_spec_3582_prompt_submission_preserves_compact_composer_footer ... ok
test interactive::ui::tests::composer::red_spec_3582_single_line_composer_uses_compact_height ... ok
test interactive::ui::tests::approval::red_spec_3582_error_attention_strip_exposes_retry_and_details_actions ... ok
test interactive::ui::tests::composer::red_spec_3582_composer_footer_uses_compact_dense_labels ... ok
test interactive::ui::tests::composer::red_spec_3582_composer_hints_expose_interrupt_retry_and_details_actions ... ok
test interactive::ui::tests::composer::red_spec_3582_composer_uses_action_chips_instead_of_instruction_sentence ... ok
test interactive::ui::tests::composer::integration_spec_3582_thinking_command_opens_overlay_through_real_input_path ... ok
test interactive::ui::tests::composer::red_spec_3582_thinking_overlay_surfaces_operator_state_and_preview ... ok
test interactive::ui::tests::detail::integration_spec_3582_memory_command_switches_detail_context_through_real_input_path ... ok
test interactive::ui::tests::detail::red_spec_3582_bare_memory_command_switches_detail_context ... ok
test interactive::ui::tests::detail::integration_spec_3582_sessions_detail_surfaces_real_session_metrics ... ok
test interactive::ui::tests::detail::red_spec_3582_cortex_detail_surfaces_operator_phase_and_reason_code ... ok
test interactive::ui::tests::detail::red_spec_3582_cortex_detail_surfaces_live_runtime_posture ... ok
test interactive::ui::tests::detail::red_spec_3582_default_layout_collapses_detail_drawer_until_requested ... ok
test interactive::ui::tests::detail::red_spec_3582_bare_sessions_and_cortex_commands_switch_detail_context ... ok
test interactive::ui::tests::detail::red_spec_3582_details_drawer_exposes_context_sections_beyond_tools ... ok
test interactive::ui::tests::detail::red_spec_3582_details_drawer_uses_tabbed_context_navigation ... ok
test interactive::ui::tests::detail::red_spec_3582_memory_detail_surfaces_degraded_state_marker ... ok
test interactive::ui::tests::detail::red_spec_3582_memory_detail_surfaces_recent_user_context ... ok
test interactive::ui::tests::detail_overlay::integration_spec_3582_ctrl_t_opens_and_escape_closes_narrow_detail_overlay ... ok
test interactive::ui::tests::detail::red_spec_3582_sessions_detail_surfaces_last_prompt_and_assistant_count ... ok
test interactive::ui::tests::detail_overlay::integration_spec_3582_escape_closes_narrow_detail_overlay_from_real_key_path ... ok
test interactive::ui::tests::detail_overlay::integration_spec_3582_narrow_detail_overlay_cycles_sections_from_real_key_path ... ok
test interactive::ui::tests::detail_overlay::red_spec_3582_narrow_detail_overlay_exposes_close_and_section_hints ... ok
test interactive::ui::tests::detail_overlay::red_spec_3582_narrow_layout_uses_detail_overlay ... ok
test interactive::ui::tests::palette::integration_spec_3582_command_palette_executes_memory_command_through_ctrl_p_path ... ok
test interactive::ui::tests::palette::integration_spec_3582_transcript_command_exports_plaintext_transcript ... ok
test interactive::ui::tests::palette::red_spec_3582_command_palette_escape_returns_focus_to_input ... ok
test interactive::ui::tests::palette::red_spec_3582_command_palette_filters_commands_from_live_input ... ok
test interactive::ui::tests::palette::red_spec_3582_command_palette_lists_common_commands_and_shortcuts ... ok
test interactive::ui::tests::palette::red_spec_3582_typing_slash_from_empty_input_opens_command_palette ... ok
test interactive::ui::tests::transcript::integration_spec_3582_prompt_submission_surfaces_last_turn_summary_card ... ok
test interactive::ui::tests::transcript::integration_spec_3582_prompt_submission_uses_compact_transcript_headers ... ok
test interactive::ui::tests::transcript::red_spec_3582_compact_transcript_keeps_multiline_assistant_output_aligned ... ok
test interactive::ui::tests::transcript::red_spec_3582_failed_run_state_card_surfaces_reason_code ... ok
test interactive::ui::tests::transcript::red_spec_3582_live_activity_surfaces_active_tool_count_and_name ... ok
test interactive::ui::tests::transcript::red_spec_3582_status_bar_surfaces_gateway_transport_posture ... ok
test interactive::ui::tests::transcript::red_spec_3582_live_activity_surfaces_operator_entity_and_artifact_kind ... ok
test interactive::ui::tests::transcript::red_spec_3582_status_bar_surfaces_local_transport_when_gateway_missing ... ok
test interactive::ui::tests::transcript::red_spec_3582_status_bar_surfaces_session_and_approval_context ... ok
test interactive::ui::tests::transcript::red_spec_3582_status_bar_surfaces_operator_phase_context ... ok
test interactive::ui::tests::transcript::red_spec_3582_streaming_run_state_card_surfaces_artifact_context ... ok
test interactive::ui::tests::transcript::red_spec_3582_thinking_run_state_card_uses_phase_specific_title ... ok
test interactive::ui::tests::transcript::red_spec_3582_streaming_turn_card_surfaces_assistant_preview ... ok
test tests::functional_apply_overlay_extends_canvas_when_needed ... ok
test tests::functional_editor_buffer_multiline_editing_and_navigation ... ok
test tests::functional_editor_view_hides_line_numbers_when_disabled ... ok
test interactive::ui::tests::transcript::red_spec_3582_transcript_shows_live_activity_summary_above_messages ... ok
test tests::functional_luma_image_renders_gradient_to_ascii ... ok
test tests::functional_operator_shell_renderer_includes_auth_rows_and_training_metrics ... ok
test tests::functional_live_shell_frame_loads_dashboard_and_training_artifacts ... ok
test tests::integration_default_theme_is_valid ... ok
test tests::integration_editor_buffer_diff_renderer_tracks_line_changes ... ok
test tests::integration_editor_view_overlay_and_diff_renderer_updates_cursor_line_only ... ok
test tests::functional_theme_from_path_loads_valid_file ... ok
test tests::integration_renderer_diff_with_overlay_updates_only_changed_lines ... ok
test tests::regression_apply_overlay_handles_unicode_content ... ok
test tests::regression_editor_delete_backward_merges_lines_without_panic ... ok
test tests::regression_luma_image_render_handles_zero_width ... ok
test tests::regression_theme_rejects_invalid_color_code ... ok
test interactive::ui::tests::transcript::red_spec_3582_transcript_surfaces_active_turn_card_with_prompt_context ... ok
test tests::renderer_outputs_only_changed_lines ... ok
test tests::spec_c01_operator_shell_renderer_outputs_expected_panel_headers ... ok
test tests::regression_live_shell_frame_handles_missing_artifacts_without_panicking ... ok
test tests::spec_c28_regression_operator_shell_auth_panel_requires_auth_mode_and_required_markers ... ok
test tests::spec_c28_regression_operator_shell_status_and_alert_panels_require_control_plane_markers ... ok
test tests::text_component_renders_with_wrap ... ok
test tests::unit_apply_overlay_replaces_existing_text ... ok
test tests::unit_editor_buffer_insert_and_delete_single_line ... ok
test tests::unit_editor_view_renders_line_numbers_and_cursor ... ok
test tests::unit_luma_image_rejects_invalid_pixel_count ... ok
test tests::unit_theme_from_json_parses_and_paints_text ... ok
test tests::wraps_long_word ... ok
test tests::wraps_text_to_width ... ok
test tests::spec_c01_live_shell_frame_reports_malformed_json_artifact_diagnostics ... ok
test tests::spec_c02_live_shell_frame_reports_malformed_jsonl_artifact_diagnostics ... ok
test interactive::ui::tests::transcript::red_spec_3582_transcript_surfaces_running_tool_card_without_opening_drawer ... ok
test interactive::ui::tests::transcript::red_spec_3582_transcript_uses_compact_headers_without_box_chrome ... ok
test interactive::gateway_runtime_tests::red_spec_3585_gateway_runtime_surfaces_supported_model_hint_for_oauth_incompatibility ... ok
test interactive::gateway_runtime_tests::integration_spec_3582_gateway_runtime_renders_phase_specific_streaming_visuals ... ok
test interactive::gateway_runtime_tests::integration_spec_3582_gateway_runtime_streams_operator_state_into_app ... ok
test interactive::gateway_runtime_tests::integration_spec_3582_gateway_runtime_ignores_keep_alive_comment_frames ... ok
test interactive::ui::tests::palette::integration_spec_3582_copy_command_writes_transcript_to_clipboard_command ... ok

test result: ok. 90 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.18s

running 24 tests
test tests::functional_spec_3474_c03_live_watch_marker_contract_is_deterministic ... ok
test tests::functional_parse_args_supports_custom_values ... ok
test tests::functional_spec_3474_c04_help_text_exposes_shell_live_watch_flags ... ok
test tests::functional_spec_c05_help_text_exposes_agent_flags ... ok
test tests::functional_spec_c05_build_agent_runtime_command_contract_is_stable ... ok
test tests::integration_spec_3474_c01_parse_args_accepts_shell_live_watch_mode_controls ... ok
test tests::functional_spec_c08_agent_runtime_command_includes_timeout_retry_flags ... ok
test tests::red_spec_3582_interactive_gateway_config_defaults_to_local_unified_bind ... ok
test tests::red_spec_3582_interactive_gateway_config_prefers_explicit_auth_token ... ok
test tests::red_spec_3585_agent_mode_defaults_to_gpt_5_3_codex_for_local_tui ... ok
test tests::red_spec_3585_interactive_mode_defaults_to_gpt_5_3_codex_for_local_tui ... ok
test tests::regression_parse_args_rejects_shell_live_state_dir_without_value ... ok
test tests::regression_compose_frame_overlays_frame_metadata ... ok
test tests::regression_parse_args_rejects_zero_frames ... ok
test tests::regression_spec_3474_c02_parse_args_rejects_shell_live_zero_iterations ... ok
test tests::regression_spec_c05_parse_args_rejects_agent_gateway_state_dir_without_value ... ok
test tests::regression_spec_c06_agent_mode_defaults_to_codex_baseline ... ok
test tests::spec_c01_parse_args_accepts_shell_mode_and_overrides ... ok
test tests::spec_c02_parse_args_rejects_shell_profile_without_value ... ok
test tests::spec_c03_parse_args_accepts_shell_live_mode_and_state_dir ... ok
test tests::spec_c05_parse_args_accepts_agent_mode_and_overrides ... ok
test tests::spec_c07_parse_args_accepts_agent_timeout_and_retry_overrides ... ok
test tests::spec_c01_shell_renderer_includes_all_operator_panels ... ok
test tests::unit_parse_args_defaults_are_stable ... ok

test result: ok. 24 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

running 6 tests
test regression_tui_demo_binary_rejects_invalid_frames_argument ... ok
test conformance_tui_agent_mode_dry_run_emits_interactive_launch_contract ... ok
test conformance_tui_shell_binary_renders_operator_panels ... ok
test functional_tui_demo_binary_renders_single_frame_without_color ... ok
test integration_tui_interactive_mode_fails_loud_without_tty ... ok
test integration_tui_demo_binary_renders_multiple_frames ... ok

test result: ok. 6 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.57s

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

This addresses the current usability gap where operators had to screenshot the TUI to preserve conversation output. The next useful slice is richer transcript readability and live execution detail, not basic command discoverability or transcript extraction.

[njfio]

Implemented the next #3582 TUI slice on 3582-tui-ux-next.

What landed:

• typing / from an empty composer now opens the command palette directly instead of inserting raw text
• added transcript actions through the real command surface:
  • /transcript exports the current transcript to .tau/tmp/transcript-latest.txt
  • /copy copies the current transcript through the configured clipboard command
• command palette defaults now surface transcript/share actions instead of hiding them behind a long command list
• transcript export/copy actions are wired through the same interactive input path as the rest of the shell
• clipboard/export behavior was split into focused modules to stay inside the refactor gate

Validation:

• cargo test -p tau-tui

This addresses the current usability gap where operators had to screenshot the TUI to preserve conversation output. The next useful slice is richer transcript readability and live execution detail, not basic command discoverability or transcript extraction.