tui-agent-activity

Metadata

Description

## Description

User reports: the TUI has a '4: Log' view (tab 4) but it 'has no direct view of the actual events' from running agents. So the pane exists — it just isn't surfacing what the in-progress agents are actually DOING.

Today: an agent is running (you can see it in the agent/task list), but you can't watch its tool calls, file edits, partial output, or decision stream from the TUI. Have to drop to `tail -f .wg/agents/agent-N/raw_stream.jsonl` in another terminal.

### Spec — what the Log view should show

For each in-progress agent (or a selectable per-agent stream):
- Live stream of events as they happen: tool calls (with args), tool results (truncated), file edits (path + lines changed), text output, model thinking blocks (when present)
- Auto-scroll on new events; pgup/pgdown / scroll wheel to scroll back; G/gg jump to bottom/top
- Per-agent filter — switch between agents to see different streams. Default: show ALL agent streams interleaved with agent-id prefix
- Persistence: when an agent finishes, the log stays viewable (read from the on-disk `raw_stream.jsonl` for that agent)
- Color-code by event type (tool call vs result vs text vs error)

### Investigation tasks

- Locate the current Log view source (`src/tui/log_pane.rs`? grep for '4: Log' or 'Log' tab key in TUI). What does it currently render?
- Where do agent events live? `.wg/agents/agent-N/raw_stream.jsonl` per the daemon — verify format and schema.
- Is there a tail-and-parse path already, or do we need to build one?

### Files likely to touch

- The existing log-pane file (find via grep)
- `src/tui/viz_viewer/state.rs` — pane state machine
- `src/tui/mod.rs` — top-level layout / tab routing

## Validation

- [ ] Failing test first (TUI test caveats apply): test_log_view_renders_agent_stream_events_for_inprogress_task
- [ ] Implementation makes test pass
- [ ] cargo build + cargo test pass with no regressions
- [ ] Manual smoke (live):
  - Start service, publish a task that takes >1 min, agent spawns
  - In wg tui, switch to '4: Log' tab → see live event stream from the spawned agent (tool calls, edits, output) populating in real time
  - Scroll back to see earlier events
  - With multiple agents alive, see all streams (per-agent prefixed) OR a per-agent picker
  - Agent finishes → log stays visible

Depends on

• done .assign-tui-agent-activity

Required by

• (none)

Log

• 2026-04-26T16:43:16.381029155+00:00 Task paused
• 2026-04-26T16:44:37.669457382+00:00 Task published
• 2026-04-26T16:45:00.756437279+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Programmer role essential for TUI feature implementation; Careful tradeoff aligns with TDD-first validation and exhaustive live smoke testing requirements.
• 2026-04-26T16:45:01.084697377+00:00 Spawned by coordinator --executor claude --model claude-opus-4-6
• 2026-04-26T16:45:08.617276550+00:00 Starting work on TUI Log view agent activity display
• 2026-04-26T16:47:26.728318436+00:00 Investigation complete: Current Log tab reads output.log via extract_enriched_text_from_log. raw_stream.jsonl exists per-agent with structured events (tool_use, tool_result, text, thinking). Plan: add structured event parsing from raw_stream.jsonl, color-coded rendering, and per-agent filtering.
• 2026-04-26T17:04:02.956930978+00:00 Validated: all 12 new agent_stream_tests pass, all existing log-tab tests pass, cargo build succeeds. 6 pre-existing failures in spawn_task/provenance unrelated to changes.
• 2026-04-26T17:04:38.386927479+00:00 Committed: 594cd3f0f — pushed to remote
• 2026-04-26T17:05:03.329444147+00:00 Task marked as done