tui-log-view-4 — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-820`
Agent identity	`f51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e`
Created	2026-04-27T17:41:36.380542857+00:00
Started	2026-04-27T18:40:57.002795920+00:00
Completed	2026-04-27T19:02:24.260902322+00:00
Tags	`eval-scheduled`
Tokens	595119 in / 30202 out

Description

Refinement on top of tui-log-view-2 (done) and tui-log-view-3 (done). User feedback after using the fixed Log view:

1. Hanging indent

Continuation lines should align under the content of the first line, not under the checkmark. Concretely:

✓ 1 #!/usr/bin/env bash
  2 # Helpers shared by smoke-gate scenarios.
  3 #

First line: <checkmark><space><content> — content starts at column 3. Continuation lines: <space><space><content> — also at column 3.

Net effect: every line of multi-line content sits at the same column. Only the first line carries the ✓ indicator; subsequent lines have 2 leading spaces of pure padding.

User quote: 'what about keeping two spaces at the beginning of the line? so we'd have checkmark space and then two spaces on each line below?'

2. Vertical gaps for legibility

Add a blank line at two specific boundaries:

Before a tool use — emphasizes 'the agent has stopped thinking and is about to act'
Before agent text — emphasizes 'the agent is starting to speak/write now'

User quote: 'maybe before tool use is the cleanest spot ... and before the agent writes text. makes sense to show they are writing something in the log.'

Do NOT add gaps between consecutive tool-result lines, between consecutive lines of agent text, or between events of the same type. Only at the transition boundaries:

prev event was agent-text, next is tool-use → blank line
prev event was tool-result, next is agent-text → blank line
prev event was agent-text, next is agent-text (continuation, same turn) → NO blank line

Files likely to touch

src/tui/viz_viewer/render.rs — Log view paragraph builders, around the spans that prepend ✓ / status indicators on the first line. Needs to switch first-line-only prefix to a hanging-indent ratatui.Wrap config OR generate the leading spaces explicitly per line.
Same render.rs, the event-stream walker — track previous event type and emit a blank Line between events when boundary is (text↔tool).

Validation

Failing test first: render Log view with a 5-line code block prefixed by ✓ → assert lines 2-5 each begin with exactly 2 spaces (hanging indent), line 1 begins with ✓ + space, and content text aligns at column 3 on every line
Failing test for gap rule: render an event stream of [agent-text, tool-use, tool-result, agent-text] → assert blank Line emitted between agent-text and tool-use, AND between tool-result and agent-text; NO blank Line between tool-use and its tool-result
Failing test that consecutive same-type events do NOT get blank lines between them: [agent-text, agent-text, agent-text] → no blank Lines inserted
Implementation makes all tests pass
cargo build + cargo test pass with no regressions
Manual smoke: open wg tui on a chat with multi-line tool results AND interleaved agent text + tool use; visually verify the indent is uniform across continuation lines, and the blank-line gaps appear at the right transition boundaries (and ONLY there)

## Description

Refinement on top of `tui-log-view-2` (done) and `tui-log-view-3` (done). User feedback after using the fixed Log view:

### 1. Hanging indent

Continuation lines should align under the content of the first line, not under the checkmark. Concretely:

```
✓ 1 #!/usr/bin/env bash
2 # Helpers shared by smoke-gate scenarios.
3 #
```

First line: `<checkmark><space><content>` — content starts at column 3.
Continuation lines: `<space><space><content>` — also at column 3.

Net effect: every line of multi-line content sits at the same column. Only the first line carries the `✓` indicator; subsequent lines have 2 leading spaces of pure padding.

User quote: 'what about keeping two spaces at the beginning of the line? so we'd have checkmark space and then two spaces on each line below?'

### 2. Vertical gaps for legibility

Add a blank line at two specific boundaries:

- **Before a tool use** — emphasizes 'the agent has stopped thinking and is about to act'
- **Before agent text** — emphasizes 'the agent is starting to speak/write now'

User quote: 'maybe before tool use is the cleanest spot ... and before the agent writes text. makes sense to show they are writing something in the log.'

Do NOT add gaps between consecutive tool-result lines, between consecutive lines of agent text, or between events of the same type. Only at the **transition boundaries**:
- prev event was agent-text, next is tool-use → blank line
- prev event was tool-result, next is agent-text → blank line
- prev event was agent-text, next is agent-text (continuation, same turn) → NO blank line

## Files likely to touch

- src/tui/viz_viewer/render.rs — Log view paragraph builders, around the spans that prepend `✓` / status indicators on the first line. Needs to switch first-line-only prefix to a hanging-indent ratatui.Wrap config OR generate the leading spaces explicitly per line.
- Same render.rs, the event-stream walker — track previous event type and emit a blank Line between events when boundary is (text↔tool).

## Validation

- [ ] Failing test first: render Log view with a 5-line code block prefixed by ✓ → assert lines 2-5 each begin with exactly 2 spaces (hanging indent), line 1 begins with ✓ + space, and content text aligns at column 3 on every line
- [ ] Failing test for gap rule: render an event stream of [agent-text, tool-use, tool-result, agent-text] → assert blank Line emitted between agent-text and tool-use, AND between tool-result and agent-text; NO blank Line between tool-use and its tool-result
- [ ] Failing test that consecutive same-type events do NOT get blank lines between them: [agent-text, agent-text, agent-text] → no blank Lines inserted
- [ ] Implementation makes all tests pass
- [ ] cargo build + cargo test pass with no regressions
- [ ] Manual smoke: open wg tui on a chat with multi-line tool results AND interleaved agent text + tool use; visually verify the indent is uniform across continuation lines, and the blank-line gaps appear at the right transition boundaries (and ONLY there)

Depends on

done .assign-tui-log-view-4

Required by

(none)

Log

2026-04-27T17:41:36.301877852+00:00 Task paused
2026-04-27T17:41:48.008951302+00:00 Task published
2026-04-27T17:42:19.857873535+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Careful Programmer has proven TUI/rendering expertise (0.72 score, 111 tasks) and Careful tradeoff prioritizes correctness—ideal for precision-critical indent/gap formatting and test-driven work.
2026-04-27T17:42:23.463642437+00:00 Spawned by coordinator --executor claude --model opus
2026-04-27T17:42:35.669423703+00:00 Starting task: hanging-indent + blank-line gaps in Log view
2026-04-27T17:49:07.647059224+00:00 Task marked as failed: Agent exited with code 1
2026-04-27T18:39:11.139741402+00:00 Reset by `wg recover` — reason: openrouter outage cleanup; everything moved to claude:opus
2026-04-27T18:39:53.599858016+00:00 Spawned by coordinator --executor claude --model opus
2026-04-27T18:40:14.825335874+00:00 Resuming from prior agent's WIP — found minimal additions to test helpers in log_render.rs but main implementation not yet done. Reading current state to plan implementation.
2026-04-27T18:40:54.096556986+00:00 Task unclaimed: agent 'agent-809' (PID 1982540) process exited
2026-04-27T18:40:57.002800739+00:00 Spawned by coordinator --executor claude --model opus
2026-04-27T19:01:37.242381793+00:00 Implementation complete. Tests pass: test_raw_mode_tool_result_uses_hanging_indent (5-line block: ✓ on first line, 2-space pad on lines 2-5, no bare result header), test_raw_mode_error_tool_result_uses_hanging_indent (✗ variant), test_raw_mode_inserts_blank_at_text_to_tool_boundary (exactly 2 blanks at category boundaries, none between tool call+result), test_raw_mode_no_blank_between_consecutive_text_events (3 text events, 0 blanks), test_raw_mode_no_blank_between_consecutive_tool_events (3 tool events, 0 blanks). Pre-existing test_raw_mode_renders_user_messages_pretty + test_raw_mode_renders_tool_calls_pretty_not_json still pass. The 7 unrelated test failures (spawn_task, provenance_coverage, test_legacy_coordinator_tab_is_always_muted) all reproduce on a clean stash, confirming they are pre-existing and not caused by these changes.
2026-04-27T19:02:17.197458359+00:00 Committed: 7a14d25c3 — pushed to origin/wg/agent-784/tui-log-view-4
2026-04-27T19:02:24.260922330+00:00 Task marked as done