Merge pull request #12 from BackendStack21/feat/tui-expand-tool-steps

feat(tui): expand tool steps via keyboard (^E) and mouse

Author: jkyberneees

internal/tui/banner.go

@@ -50,6 +50,7 @@ func welcome(th theme, width int, cwd string) string {
 		{"⏎ send", "^J newline  ·  ^T toggle thinking"},
 		{"^L clear", "↑/↓ scroll  ·  PgUp/PgDn page  ·  ^C quit"},
 		{"approvals", "a approve  ·  d deny  ·  t trust"},
+		{"tool steps", "^E expand the last step  ·  --mouse to click-expand"},
 	}
 	const keyW = 11
 	for _, t := range tips {

internal/tui/commands.go

@@ -174,8 +174,10 @@ func (m *Model) showHelp() {
 		{"^O", "switch model"},
 		{"^T", "toggle extended thinking"},
 		{"^L", "clear the conversation"},
+		{"^E", "expand the last tool step"},
 		{"esc", "cancel the running turn"},
 		{"^C", "quit"},
+		{"--mouse", "click tool rows to expand"},
 	} {
 		b.WriteString("\n" + th.tipKey.Render(padRight(k[0], keyW)) + " " + th.tipText.Render(k[1]))
 	}

internal/tui/final_gaps_test.go

@@ -37,13 +37,13 @@ func TestHeaderThinkAndSandbox(t *testing.T) {
 func TestRenderNoteAndStepDefaultIcon(t *testing.T) {
 	m := wired(t)
 	// roleNote rendering.
-	out := m.renderMessage(message{role: roleNote, content: "a note"})
+	out, _ := m.renderMessage(message{role: roleNote, content: "a note"}, 0, 0)
 	if !strings.Contains(plain(out), "a note") {
 		t.Error("note message not rendered")
 	}
 	// A finalized assistant message with an unfinished step uses the ▸ icon.
 	msg := message{role: roleAsst, content: "done", steps: []step{{name: "shell", done: false}}}
-	out = m.renderMessage(msg)
+	out, _ = m.renderMessage(msg, 0, 0)
 	if !strings.Contains(plain(out), "shell") {
 		t.Error("step not rendered")
 	}

internal/tui/integration_test.go

@@ -153,6 +153,8 @@ func key(s string) tea.KeyMsg {
 		return tea.KeyMsg{Type: tea.KeyCtrlL}
 	case "ctrl+j":
 		return tea.KeyMsg{Type: tea.KeyCtrlJ}
+	case "ctrl+e":
+		return tea.KeyMsg{Type: tea.KeyCtrlE}
 	default:
 		return tea.KeyMsg{Type: tea.KeyRunes, Runes: []rune(s)}
 	}

internal/tui/model.go

@@ -36,6 +36,15 @@ type step struct {
 	isErr    bool     // the result reads as a failure (tints the status glyph red)
 	subagent bool     // this call delegates to a sub-agent (renders its log tree)
 	logs     []string // nested sub-agent activity, from subagent_log events
+	expanded bool     // user has expanded this step to show full output/logs
+}
+
+// stepRef maps a rendered transcript line to a specific step for mouse
+// hit-testing.
+type stepRef struct {
+	msgIdx  int
+	stepIdx int
+	line    int
 }
 
 // turnStats is the telemetry of one finalized assistant turn, captured from the
@@ -56,6 +65,7 @@ type message struct {
 	role      role
 	content   string // raw text/markdown
 	rendered  string // cached glamour render (assistant, finalized)
+	thinking  string // captured reasoning for this turn (finalized)
 	steps     []step
 	streaming bool
 	stats     *turnStats // finalized-turn telemetry; nil while streaming / for history
@@ -131,8 +141,10 @@ type Model struct {
 	gradRuleW int
 	logoCache string // cached gradient logo (width-independent)
 
-	convPrefix string // cached rendering of the finalized transcript prefix
-	convCount  int    // messages the prefix covers (-1 = invalidated)
+	convPrefix     string    // cached rendering of the finalized transcript prefix
+	convPrefixRefs []stepRef // step header line index for the cached prefix
+	convCount      int       // messages the prefix covers (-1 = invalidated)
+	stepLineIndex  []stepRef // full transcript step index for mouse hit-testing
 }
 
 // acMode selects what the completion popup is completing.
@@ -292,6 +304,18 @@ func (m *Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		return m, nil
 
 	case tea.MouseMsg:
+		if msg.Action == tea.MouseActionPress && msg.Button == tea.MouseButtonLeft && m.panel == panelNone && !m.ac.open {
+			// Viewport content begins below the header (2 rows).
+			top := 2
+			if msg.Y >= top && msg.Y < top+m.vp.Height {
+				line := msg.Y - top + m.vp.YOffset
+				if msgIdx, stepIdx, ok := m.stepAtLine(line); ok {
+					m.toggleStep(msgIdx, stepIdx)
+					m.refresh()
+					return m, nil
+				}
+			}
+		}
 		var cmd tea.Cmd
 		m.vp, cmd = m.vp.Update(msg)
 		return m, cmd
@@ -386,6 +410,10 @@ func (m *Model) handleKey(msg tea.KeyMsg) (tea.Model, tea.Cmd) {
 			m.clearConversation()
 		}
 		return m, nil
+	case "ctrl+e":
+		m.toggleRecentStep()
+		m.refresh()
+		return m, nil
 	case "up", "ctrl+p":
 		// Scroll the transcript when the cursor is already at the top line of
 		// the input; otherwise let the textarea move the cursor up.
@@ -559,6 +587,8 @@ func (m *Model) clearConversation() {
 	m.msgs = nil
 	m.curIdx = -1
 	m.convCount = -1 // transcript replaced — drop the cached prefix
+	m.convPrefixRefs = nil
+	m.stepLineIndex = nil
 	m.turnStats = nil
 	m.toolTotal = 0
 	m.sessionStart = time.Time{}
@@ -701,6 +731,7 @@ func (m *Model) finalize() {
 	if i := m.cur(); i >= 0 {
 		m.msgs[i].streaming = false
 		m.msgs[i].rendered = m.render(m.msgs[i].content)
+		m.msgs[i].thinking = m.thinking.String()
 	}
 	m.curIdx = -1
 	m.thinking.Reset()
@@ -1117,3 +1148,47 @@ func truncate(s string, n int) string {
 	}
 	return string(r[:n-1]) + "…"
 }
+
+// toggleStep flips the expanded state of a step and invalidates the transcript
+// prefix cache so the re-render picks it up.
+func (m *Model) toggleStep(msgIdx, stepIdx int) {
+	if msgIdx < 0 || msgIdx >= len(m.msgs) {
+		return
+	}
+	steps := m.msgs[msgIdx].steps
+	if stepIdx < 0 || stepIdx >= len(steps) {
+		return
+	}
+	steps[stepIdx].expanded = !steps[stepIdx].expanded
+	m.convCount = -1
+}
+
+// toggleRecentStep expands/collapses the last step of the most recent message
+// that has steps. Bound to the "e" key.
+func (m *Model) toggleRecentStep() {
+	for i := len(m.msgs) - 1; i >= 0; i-- {
+		if len(m.msgs[i].steps) > 0 {
+			m.toggleStep(i, len(m.msgs[i].steps)-1)
+			return
+		}
+	}
+}
+
+// stepAtLine maps a viewport content line to the nearest step header at or
+// above it. Used for mouse click-to-expand.
+func (m *Model) stepAtLine(line int) (msgIdx, stepIdx int, ok bool) {
+	if len(m.stepLineIndex) == 0 {
+		return
+	}
+	var ref *stepRef
+	for i := range m.stepLineIndex {
+		if m.stepLineIndex[i].line > line {
+			break
+		}
+		ref = &m.stepLineIndex[i]
+	}
+	if ref == nil {
+		return
+	}
+	return ref.msgIdx, ref.stepIdx, true
+}

internal/tui/steps_test.go

@@ -109,8 +109,8 @@ func TestSubagentLogNesting(t *testing.T) {
 	}
 }
 
-// TestRenderStepsSubagentAndError exercises the enriched step rendering: a
-// sub-agent label, a nested log tree, and an error-tinted result.
+// TestRenderStepsSubagentAndError exercises the one-line step summary: a
+// sub-agent label, a compact result arrow, and an error-tinted result.
 func TestRenderStepsSubagentAndError(t *testing.T) {
 	m := newTestModel()
 	msg := message{
@@ -123,21 +123,155 @@ func TestRenderStepsSubagentAndError(t *testing.T) {
 				result: "exit status 1\nFAIL"},
 		},
 	}
-	out := plain(m.renderSteps(msg))
-	for _, want := range []string{"sub-agent", "⎿", "explorer", "✗", "exit status 1"} {
-		if !strings.Contains(out, want) {
-			t.Errorf("renderSteps missing %q in:\n%s", want, out)
+	out, _ := m.renderSteps(msg, 0, 0)
+	plainOut := plain(out)
+	for _, want := range []string{"sub-agent", "delegate_task", "explore", "→ done", "✗", "shell", "go test", "→ exit status 1", "▶"} {
+		if !strings.Contains(plainOut, want) {
+			t.Errorf("renderSteps missing %q in:\n%s", want, plainOut)
 		}
 	}
+	// Nested logs are still captured but no longer expanded in the default render.
+	if strings.Contains(plainOut, "⎿") || strings.Contains(plainOut, "explorer") {
+		t.Errorf("renderSteps should not expand nested logs in one-line mode:\n%s", plainOut)
+	}
 
 	// Streaming turn: a not-done step renders the live spinner; a not-done step
 	// in a finalized turn renders the pending glyph. Also drive the narrow-width
 	// budget floor.
 	m.vp.Width = 8
-	if s := m.renderSteps(message{streaming: true, steps: []step{{name: "read", arg: "x"}}}); s == "" {
+	if s, _ := m.renderSteps(message{streaming: true, steps: []step{{name: "read", arg: "x"}}}, 0, 0); s == "" {
 		t.Error("streaming step rendered empty")
 	}
-	if s := m.renderSteps(message{streaming: false, steps: []step{{name: "read"}}}); !strings.Contains(plain(s), "▸") {
+	if s, _ := m.renderSteps(message{streaming: false, steps: []step{{name: "read"}}}, 0, 0); !strings.Contains(plain(s), "▸") {
 		t.Errorf("pending step missing ▸ glyph: %q", plain(s))
 	}
 }
+
+func TestRenderStepsExpanded(t *testing.T) {
+	m := newTestModel()
+	msg := message{role: roleAsst, streaming: false, steps: []step{
+		{name: "shell", arg: "go test", done: true, isErr: true, expanded: true,
+			result: "exit status 1\nFAIL"},
+	}}
+	out, refs := m.renderSteps(msg, 5, 0)
+	plainOut := plain(out)
+	for _, want := range []string{"▼", "shell", "go test", "exit status 1", "FAIL"} {
+		if !strings.Contains(plainOut, want) {
+			t.Errorf("expanded step missing %q in:\n%s", want, plainOut)
+		}
+	}
+	if strings.Contains(plainOut, "▶") {
+		t.Errorf("expanded step should not show ▶ in:\n%s", plainOut)
+	}
+	if len(refs) != 1 || refs[0].line != 5 {
+		t.Errorf("unexpected step refs: %+v", refs)
+	}
+}
+
+func TestResultOneLiner(t *testing.T) {
+	if got := resultOneLiner("first\nsecond", 10); got != "first" {
+		t.Errorf("resultOneLiner first line: %q", got)
+	}
+	if got := resultOneLiner("\n  \nsecond", 8); got != "second" {
+		t.Errorf("resultOneLiner skips blanks: %q", got)
+	}
+	if got := resultOneLiner("", 8); got != "" {
+		t.Errorf("resultOneLiner empty: %q", got)
+	}
+	if got := resultOneLiner("a very long first meaningful line", 10); got != "a very lo…" {
+		t.Errorf("resultOneLiner truncation: %q", got)
+	}
+}
+
+func TestToggleStep(t *testing.T) {
+	m := newTestModel()
+	m.msgs = append(m.msgs, message{role: roleAsst, steps: []step{
+		{name: "shell", arg: "go test", done: true, result: "exit status 1\nFAIL"},
+	}})
+	m.toggleStep(0, 0)
+	if !m.msgs[0].steps[0].expanded {
+		t.Error("toggleStep did not expand the step")
+	}
+	if m.convCount != -1 {
+		t.Error("toggleStep did not invalidate the transcript cache")
+	}
+	out := plain(m.conversation())
+	if !strings.Contains(out, "▼") || !strings.Contains(out, "FAIL") {
+		t.Errorf("expanded step not rendered:\n%s", out)
+	}
+}
+
+func TestToggleRecentStep(t *testing.T) {
+	m := newTestModel()
+	m.msgs = append(m.msgs,
+		message{role: roleAsst, steps: []step{{name: "read", done: true, result: "ok"}}},
+		message{role: roleAsst, steps: []step{{name: "shell", done: true, result: "done"}}},
+	)
+	m.toggleRecentStep()
+	if !m.msgs[1].steps[0].expanded {
+		t.Error("toggleRecentStep did not expand the most recent step")
+	}
+	if m.msgs[0].steps[0].expanded {
+		t.Error("toggleRecentStep expanded the wrong step")
+	}
+}
+
+func TestKeyCtrlEExpandsStep(t *testing.T) {
+	m := newTestModel()
+	m.ta.Focus()
+	m.msgs = append(m.msgs, message{role: roleAsst, steps: []step{{name: "shell", done: true, result: "ok"}}})
+
+	// Pressing Ctrl+E should expand the most recent step, even while typing.
+	m.ta.SetValue("hello")
+	m.Update(key("ctrl+e"))
+	if !m.msgs[0].steps[0].expanded {
+		t.Error("Ctrl+E should expand the recent step")
+	}
+	if got := m.ta.Value(); got != "hello" {
+		t.Errorf("Ctrl+E should not alter the input, got %q", got)
+	}
+}
+
+func TestKeyETypesLetter(t *testing.T) {
+	m := newTestModel()
+	m.ta.Focus()
+
+	// The plain 'e' key must always type the letter, even at the start.
+	m.Update(key("e"))
+	if got := m.ta.Value(); got != "e" {
+		t.Errorf("pressing 'e' should type the letter, got %q", got)
+	}
+	m.Update(key("e"))
+	if got := m.ta.Value(); got != "ee" {
+		t.Errorf("pressing 'e' again should append the letter, got %q", got)
+	}
+}
+
+func TestStepLineIndex(t *testing.T) {
+	m := newTestModel()
+	m.msgs = append(m.msgs,
+		message{role: roleAsst, steps: []step{{name: "read", done: true, result: "ok"}}},
+		message{role: roleAsst, steps: []step{{name: "shell", done: true, result: "done"}}},
+	)
+	_ = m.conversation()
+	if len(m.stepLineIndex) != 2 {
+		t.Fatalf("expected 2 step refs, got %+v", m.stepLineIndex)
+	}
+	if m.stepLineIndex[0].line >= m.stepLineIndex[1].line {
+		t.Errorf("step refs not in ascending order: %+v", m.stepLineIndex)
+	}
+	msgIdx, stepIdx, ok := m.stepAtLine(m.stepLineIndex[0].line)
+	if !ok || msgIdx != 0 || stepIdx != 0 {
+		t.Errorf("stepAtLine first header: got %d,%d,%v", msgIdx, stepIdx, ok)
+	}
+	msgIdx, stepIdx, ok = m.stepAtLine(m.stepLineIndex[1].line)
+	if !ok || msgIdx != 1 || stepIdx != 0 {
+		t.Errorf("stepAtLine second header: got %d,%d,%v", msgIdx, stepIdx, ok)
+	}
+	// A line between the two headers still maps to the first step.
+	mid := (m.stepLineIndex[0].line + m.stepLineIndex[1].line) / 2
+	msgIdx, stepIdx, ok = m.stepAtLine(mid)
+	if !ok || msgIdx != 0 || stepIdx != 0 {
+		t.Errorf("stepAtLine between headers: got %d,%d,%v", msgIdx, stepIdx, ok)
+	}
+}

internal/tui/thinking_test.go

@@ -35,3 +35,49 @@ func TestThinkingCap(t *testing.T) {
 		t.Errorf("latest thinking not retained: %q", m.thinking.String())
 	}
 }
+
+// TestThinkingRendersAboveTools verifies that live reasoning appears at the top
+// of the assistant body, before any tool summaries.
+func TestThinkingRendersAboveTools(t *testing.T) {
+	m := newTestModel()
+	m.msgs = append(m.msgs, message{role: roleAsst, streaming: true})
+	m.curIdx = 0
+	m.busy = true
+
+	m.handleEvent(client.Event{Type: "thinking", Content: "I need to check the file."})
+	m.handleEvent(client.Event{Type: "tool_call", Name: "read_file", Data: `{"path":"main.go"}`})
+	m.handleEvent(client.Event{Type: "tool_result", Name: "read_file", Data: "package main\n"})
+
+	rendered, _ := m.renderMessage(m.msgs[0], 0, 0)
+	out := plain(rendered)
+	thinkIdx := strings.Index(out, "I need to check the file")
+	toolIdx := strings.Index(out, "read_file")
+	if thinkIdx < 0 || toolIdx < 0 || thinkIdx > toolIdx {
+		t.Errorf("thinking should appear before tools in:\n%s", out)
+	}
+}
+
+// TestThinkingCapturedOnFinalize verifies that reasoning is stored in the
+// assistant message and still renders above the final response after the turn
+// ends.
+func TestThinkingCapturedOnFinalize(t *testing.T) {
+	m := newTestModel()
+	m.msgs = append(m.msgs, message{role: roleAsst, streaming: true})
+	m.curIdx = 0
+	m.busy = true
+
+	m.handleEvent(client.Event{Type: "thinking", Content: "planning the response"})
+	m.handleEvent(client.Event{Type: "token", Content: "hello"})
+	m.handleEvent(client.Event{Type: "done", Latency: 0.5, ContextTokens: 10, OutputTokens: 1})
+
+	if m.msgs[0].thinking != "planning the response" {
+		t.Errorf("thinking not captured: %q", m.msgs[0].thinking)
+	}
+	rendered, _ := m.renderMessage(m.msgs[0], 0, 0)
+	out := plain(rendered)
+	thinkIdx := strings.Index(out, "planning the response")
+	respIdx := strings.Index(out, "hello")
+	if thinkIdx < 0 || respIdx < 0 || thinkIdx > respIdx {
+		t.Errorf("thinking should appear above response in finalized message:\n%s", out)
+	}
+}