feat: back-port RichUIHandler output improvements to InteractiveUIHandler

[christophevg-agent]

Summary

Back-port RichUIHandler's output improvements into InteractiveUIHandler, producing a single unified interactive handler that combines RichUIHandler's stable append-only output with the old InteractiveUIHandler's prompt_toolkit-based input handling.

This PR contains the design and consensus documents only — implementation will follow in subsequent commits.

Background

Three domain reviews were conducted:

• functional-analyst: 35+ method merge plan, LiveDisplay usage audit, test strategy, 14 acceptance criteria
• api-architect: Protocol/Bridge/Batch unchanged, LiveDisplay safe to remove, __init__ params to preserve, found an f-string bug in output_stats
• ui-ux-designer: Append-only output is the right direction, flagged 5 blockers (stubbed methods must be implemented), long-value guard, processing feedback concern

Owner corrections were applied as binding decisions (see reporting/ui-back-port/consensus.md).

Confirmed Design

• Base: RichUIHandler (stable, append-only console.print, no Live region)
• Input: prompt_toolkit with lazy PromptSession (created on first get_input, not in __init__)
• Input rendering: erase_when_done=True → input erased → output_prompt renders in Panel
• Processing feedback: rich.status manual start/stop — stop on first token
• Removed: LiveDisplay (spinner.py), _exit_live, _ensure_live, state flags, spinner
• Added back: output_step_title, output_command_result, output_content, confirm_approval, agent_spawned/agent_finished, _print_wrapped, shutdown — adapted to stable console.print
• Preserved from old: FileHistory, multi-line Esc+Enter, set_input_messages, history_file, wrap_width, show_* flags, show_stats=True default
• Fixed: output_stats f-string bug (missing f prefix in RichUIHandler)
• Guard: _format_tool_details caps values > 60 chars, suppresses content/old_string/new_string for write/update

Owner Corrections (binding)

1. start() signature: KEEP title/version kwargs — external consumers (yoker-assistant) use them.
2. output_prompt: KEEP as a SEPARATE method — not absorbed into get_input. yoker-assistant gets prompts from email (MCP), not terminal input.
3. Processing feedback: Use rich.status with manual start/stop (not context manager) — stop when first response token arrives. NOT a LiveDisplay grid.
4. Visual formatting: Preserve all of RichUIHandler's output improvements (inline tool args, result size, prompt Panel, tools list in banner, grey74 thinking).

Changes

• analysis/ui-back-port-design.md — functional-analyst design
• analysis/ui-back-port-api-review.md — api-architect review
• analysis/ui-back-port-ux-review.md — ui-ux-designer review
• reporting/ui-back-port/consensus.md — consensus document with owner corrections and 14 acceptance criteria

Test Plan

• Implementation follows the 14 acceptance criteria in reporting/ui-back-port/consensus.md
• make check passes
• Manual REPL smoke test: input → Panel → Processing... → response streams

🤖 Implemented together with a coding agent.

[christophevg-agent]

Implementation Plan — Back-port RichUIHandler output improvements to InteractiveUIHandler

Confirmed Design (with owner corrections)

Base: RichUIHandler (from ../yoker-assistant/src/yoker_assistant/handler.py) — stable, append-only console.print, no Live region, no spinner, no state flags.

Input: prompt_toolkit with lazy PromptSession (created on first get_input call, not in __init__). Key bindings, FileHistory, InMemoryHistory, multiline Esc+Enter, set_input_messages all preserved.

Input rendering: erase_when_done=True on get_input/get_secret_input → input erased after Enter → output_prompt renders it in a styled Panel. output_prompt stays a separate method (not absorbed into get_input) — external consumers like yoker-assistant call it independently for prompts from non-terminal sources (email/MCP).

Processing feedback: rich.status with manual start()/stop() — start when agent begins processing, stop when first response token arrives. NOT a LiveDisplay grid, just a single status line.

Removed: LiveDisplay (src/yoker/ui/spinner.py), _exit_live, _ensure_live, _end_turn, state flags (_thinking_shown, _content_shown, _streaming_*).

Added back from old InteractiveUIHandler (adapted to stable console.print, no _exit_live): output_step_title, output_command_result, output_content, confirm_approval, agent_spawned, agent_finished, _print_wrapped, shutdown.

Preserved from RichUIHandler: inline tool args (key=value), result size (N chars), output_prompt Panel, tools list in banner, grey74 thinking with blank-line framing.

Preserved from old InteractiveUIHandler: FileHistory, InMemoryHistory, multiline Esc+Enter, set_input_messages, history_file param, wrap_width param, show_* flags, show_stats=True default.

Preserved per owner correction: start() title/version kwargs — external consumers (yoker-assistant) use them.

Fixed: output_stats f-string bug in RichUIHandler (missing f prefix).

Guard: _format_tool_details caps values > 60 chars, suppresses content/old_string/new_string for write/update tools.

Method-by-method plan (35+ methods)

Method	Action	Notes
__init__	Merge	RichUIHandler base + old params (history_file, wrap_width, show_* flags, show_stats=True, show_time=False). Lazy PromptSession: self._session = None
_get_or_create_session	New	Lazy PromptSession builder — called on first get_input/get_secret_input. Key bindings + history set up here
_create_session	Adapt	From old InteractiveUIHandler — KeyBindings, FileHistory/InMemoryHistory, multiline
start	Merge	RichUIHandler banner (tools list) + KEEP title/version kwargs
shutdown	Add back	"Goodbye!" print (no _exit_live)
get_input	Add	prompt_toolkit with erase_when_done=True → then output_prompt(result)
get_secret_input	Add	prompt_toolkit with erase_when_done=True + is_password=True (no output_prompt for secrets)
set_input_messages	Add	Predefined input (no session needed — short-circuits before session creation)
output_prompt	Keep	Separate method — styled Panel rendering
output_info	Keep	Plain console.print
output_thinking / stream_thinking	Keep	grey74, inline, blank-line framing
start_content_stream / stream_content / end_content_stream	Keep	Append-only console.print(chunk, end="")
output_tool_call	Keep	⏺ name(key=value, ...) inline args
output_tool_result	Keep	✓ Success (N chars)
output_tool_content	Keep	From RichUIHandler
output_step_title	Add back	Bootstrap wizard step headers
output_command_result	Add back	Slash command output
output_content	Add back	Content display
output_stats	Fix + Keep	Fix f-string bug, keep RichUIHandler format
output_error	Keep	Styled error output
confirm_approval	Add back	Protected files approval (from old handler, no _exit_live)
agent_spawned / agent_finished	Add back	Multi-agent lifecycle (2 lines each)
_print_wrapped	Add back	Width-based wrapping
_format_tool_details	Keep + Guard	Cap values > 60 chars, suppress content/old_string/new_string
_ensure_live / _exit_live / _end_turn	Remove	LiveDisplay machinery
_thinking_shown / _content_shown / _streaming_*	Remove	State flags
Processing status	New	rich.status manual start/stop — start on processing begin, stop on first token

Files

Modified:

• src/yoker/ui/interactive.py — rewritten (merge of both handlers)
• src/yoker/ui/__init__.py — remove LiveDisplay/live_display exports
• tests/test_ui/test_interactive.py — rewrite (console-capture instead of Live-state assertions, lazy-session tests)
• tests/test_ui/test_handler.py — trim LiveDisplay imports
• tests/test_bootstrap/test_history_security.py — call _get_or_create_session() before inspecting .history
• tests/test_ui/test_confirm_approval.py — widen PromptSession patch for lazy init

Deleted:

• src/yoker/ui/spinner.py — LiveDisplay
• tests/test_ui/test_spinner.py — spinner tests
• tests/events/test_spinner.py — spinner event tests

Unchanged:

• src/yoker/ui/handler.py — Protocol
• src/yoker/ui/bridge.py — UIBridge
• src/yoker/ui/batch.py — BatchUIHandler
• src/yoker/cli/chat.py — CLI wiring
• src/yoker/cli/init.py — init wiring
• examples/custom_handler.py — PrintUIHandler

Acceptance Criteria (14)

1. LiveDisplay (spinner.py) removed — no Live region, no _exit_live/_ensure_live
2. Lazy PromptSession — created on first get_input, not in init
3. erase_when_done=True on get_input/get_secret_input
4. output_prompt renders input in styled Panel after erase
5. Processing feedback via rich.status (manual start/stop, stop on first token)
6. Inline tool args (key=value) in output_tool_call
7. Result size (N chars) in output_tool_result
8. Tools list in welcome banner
9. All stubbed methods implemented: output_command_result, output_step_title, get_secret_input, confirm_approval
10. FileHistory, multi-line, agent_spawned/agent_finished, shutdown preserved
11. start() preserves title/version kwargs
12. output_prompt is a separate method (not absorbed into get_input)
13. make check passes
14. Manual REPL smoke test: input → Panel → Processing... → response streams

Open questions (from functional-analyst, with sensible defaults)

1. Skip empty-input Panel? Default: yes — if user just presses Enter, don't render an empty Panel.
2. Stats format? Default: keep RichUIHandler's simpler format (consistent with the simplification goal).
3. Banner hint (use /think ...)? Default: keep it (useful info, no harm).
4. confirm_approval test patch shape? Implementation detail — fixture-based (cleaner).
5. LiveDisplay removal from yoker.ui exports? Clean removal (pre-1.0, no deprecation shim needed).

These defaults will be applied unless you object.

---

Waiting for your approval before proceeding with implementation.

[christophevg]

Approved, proceed 👍

[christophevg-agent]

Addressed the consolidated PR #54 review feedback. Commit 525c5ae on feature/ui-back-port:

1. testing-engineer's 5 test gaps — all closed

Five new tests added to tests/test_ui/test_interactive.py:

• test_get_input_erase_when_done_toggle — erase_when_day toggle behavior
• test_get_input_erase_when_done_reset_after_exception — erase_when_done is reset even when an exception escapes get_input
• test_rich_status_started_and_stopped — rich.status instance is started and stopped cleanly
• test_status_stops_on_first_chunk — status is stopped on the first streamed chunk
• test_get_secret_input_predefined_input_path — get_secret_input returns the predefined input without prompting

2. code-reviewer M1 (erase_when_day try/finally) — implemented

get_input and get_secret_input now wrap the prompt loop in try/finally that resets erase_when_done = False on the session, so the audit-trail worry for confirm_approval (which needs erase_when_done left at its default) is closed even when an exception escapes the prompt loop.

3. code-reviewer M3 + L3 (dead code + wrap_width) — deleted

Removed the unused helpers _capitalize, _extract_filename, and _print_wrapped, plus the unused wrap_width parameter that was threaded down to them.

4. functional-analyst obs #1 (is_password reset on interrupt) — folded into finally

The is_password flag reset is now folded into the same finally block in get_secret_input, so the terminal is returned to normal rendering even if the user interrupts password entry (Ctrl+C / EOF).

5. Deliberately skipped (non-blocking)

• M2 (prompt_async unification) — deferred; not required for this cycle
• L1, L2, L4 — minor style nits, deferred

6. Re-review outcome

The scoped re-review (Stage a + Stage c testing + Stage c code) all approved. make check is green — 2187 tests passed.

[christophevg-agent]

Implementation complete. The UI back-port merges RichUIHandler's stable append-only output with the old InteractiveUIHandler's input machinery. Key features: lazy PromptSession, erase_when_done input-erased-then-Panel, rich.status manual start/stop processing feedback, all RichUIHandler visual formatting preserved (inline tool args, result size, prompt Panel, tools banner). LiveDisplay removed (spinner.py deleted).

Passed the full review cycle (Phase 5.6):

• functional-analyst: all 14 acceptance criteria met, all 5 owner corrections honored
• ui-ux-designer: all 5 UX blockers resolved
• testing-engineer: all 5 test gaps closed
• code-reviewer: M1/M3/L3 fixed, 5/5 maintainability
• make check: 2187 passed, 8 skipped

Ready for your review.