Fix 100% one-shot rate for Gemini/Mistral/Kiro/Goose

mac/Sources/CodeBurnMenubar/Views/HeatmapSection.swift

@@ -97,22 +97,25 @@ private struct InsightPillSwitcher: View {
     let visibleModes: [InsightMode]
 
     var body: some View {
-        HStack(spacing: 4) {
-            ForEach(visibleModes) { mode in
-                Button {
-                    selected = mode
-                } label: {
-                    Text(mode.rawValue)
-                        .font(.system(size: 11, weight: .medium))
-                        .foregroundStyle(selected == mode ? AnyShapeStyle(.white) : AnyShapeStyle(.secondary))
-                        .padding(.horizontal, 10)
-                        .padding(.vertical, 4)
-                        .background(
-                            RoundedRectangle(cornerRadius: 6)
-                                .fill(selected == mode ? AnyShapeStyle(Theme.brandAccent) : AnyShapeStyle(Color.secondary.opacity(0.10)))
-                        )
+        ScrollView(.horizontal, showsIndicators: false) {
+            HStack(spacing: 4) {
+                ForEach(visibleModes) { mode in
+                    Button {
+                        selected = mode
+                    } label: {
+                        Text(mode.rawValue)
+                            .font(.system(size: 11, weight: .medium))
+                            .fixedSize()
+                            .foregroundStyle(selected == mode ? AnyShapeStyle(.white) : AnyShapeStyle(.secondary))
+                            .padding(.horizontal, 10)
+                            .padding(.vertical, 4)
+                            .background(
+                                RoundedRectangle(cornerRadius: 6)
+                                    .fill(selected == mode ? AnyShapeStyle(Theme.brandAccent) : AnyShapeStyle(Color.secondary.opacity(0.10)))
+                            )
+                    }
+                    .buttonStyle(.plain)
                 }
-                .buttonStyle(.plain)
             }
         }
     }

src/classifier.ts

@@ -154,13 +154,22 @@ function classifyConversation(userMessage: string): TaskCategory {
 }
 
 function countRetries(turn: ParsedTurn): number {
+  const steps: string[][] = []
+  for (const call of turn.assistantCalls) {
+    if (call.toolSequence && call.toolSequence.length > 0) {
+      steps.push(...call.toolSequence)
+    } else if (call.tools.length > 0) {
+      steps.push(call.tools)
+    }
+  }
+
   let sawEditBeforeBash = false
   let sawBashAfterEdit = false
   let retries = 0
 
-  for (const call of turn.assistantCalls) {
-    const hasEdit = call.tools.some(t => EDIT_TOOLS.has(t))
-    const hasBash = call.tools.some(t => BASH_TOOLS.has(t))
+  for (const tools of steps) {
+    const hasEdit = tools.some(t => EDIT_TOOLS.has(t))
+    const hasBash = tools.some(t => BASH_TOOLS.has(t))
 
     if (hasEdit) {
       if (sawBashAfterEdit) retries++

src/parser.ts

@@ -1560,6 +1560,7 @@ function providerCallToCachedTurn(call: ParsedProviderCall): CachedTurn {
       deduplicationKey: call.deduplicationKey,
       project: call.project,
       projectPath: call.projectPath,
+      toolSequence: call.toolSequence,
     }],
   }
 }
@@ -1597,6 +1598,7 @@ function cachedCallToApiCall(call: CachedCall): ParsedApiCall {
     bashCommands: call.bashCommands,
     deduplicationKey: call.deduplicationKey,
     cacheCreationOneHourTokens: u.cacheCreationOneHourTokens || undefined,
+    toolSequence: call.toolSequence,
   }
 }
 

src/providers/gemini.ts

@@ -79,6 +79,7 @@ function parseSession(data: GeminiSession, seenKeys: Set<string>): ParsedProvide
   let totalThoughts = 0
   const allTools: string[] = []
   const bashCommands: string[] = []
+  const toolSequence: string[][] = []
   let model = ''
 
   for (const msg of geminiMessages) {
@@ -90,13 +91,16 @@ function parseSession(data: GeminiSession, seenKeys: Set<string>): ParsedProvide
     if (msg.model && !model) model = msg.model
 
     if (msg.toolCalls) {
+      const msgTools: string[] = []
       for (const tc of msg.toolCalls) {
         const mapped = toolNameMap[tc.displayName ?? ''] ?? toolNameMap[tc.name] ?? tc.displayName ?? tc.name
         allTools.push(mapped)
+        msgTools.push(mapped)
         if (mapped === 'Bash' && tc.args && typeof tc.args.command === 'string') {
           bashCommands.push(...extractBashCommands(tc.args.command))
         }
       }
+      if (msgTools.length > 0) toolSequence.push(msgTools)
     }
   }
 
@@ -137,6 +141,7 @@ function parseSession(data: GeminiSession, seenKeys: Set<string>): ParsedProvide
     costUSD,
     tools: [...new Set(allTools)],
     bashCommands: [...new Set(bashCommands)],
+    toolSequence: toolSequence.length > 1 ? toolSequence : undefined,
     timestamp: tsDate.toISOString(),
     speed: 'standard',
     deduplicationKey: dedupKey,

src/providers/goose.ts

@@ -80,14 +80,15 @@ function parseModelConfig(raw: string | null): ModelConfig {
   }
 }
 
-function extractToolsFromMessages(db: SqliteDatabase, sessionId: string): { tools: string[]; bashCommands: string[] } {
+function extractToolsFromMessages(db: SqliteDatabase, sessionId: string): { tools: string[]; bashCommands: string[]; toolSequence: string[][] } {
   const tools: string[] = []
   const bashCommands: string[] = []
   const seen = new Set<string>()
+  const toolSequence: string[][] = []
 
   try {
     const rows = db.query<{ content_json: Uint8Array | string }>(
-      "SELECT CAST(content_json AS BLOB) AS content_json FROM messages WHERE session_id = ? AND role = 'assistant' AND content_json LIKE '%toolRequest%'",
+      "SELECT CAST(content_json AS BLOB) AS content_json FROM messages WHERE session_id = ? AND role = 'assistant' AND content_json LIKE '%toolRequest%' ORDER BY created_timestamp ASC",
       [sessionId],
     )
 
@@ -98,6 +99,7 @@ function extractToolsFromMessages(db: SqliteDatabase, sessionId: string): { tool
       } catch {
         continue
       }
+      const msgTools: string[] = []
       for (const item of items) {
         if (item.type !== 'toolRequest') continue
         const rawName = item.toolCall?.value?.name ?? ''
@@ -107,6 +109,7 @@ function extractToolsFromMessages(db: SqliteDatabase, sessionId: string): { tool
           seen.add(mapped)
           tools.push(mapped)
         }
+        msgTools.push(mapped)
         if (mapped === 'Bash') {
           const cmd = item.toolCall?.value?.arguments?.command
           if (typeof cmd === 'string') {
@@ -116,10 +119,11 @@ function extractToolsFromMessages(db: SqliteDatabase, sessionId: string): { tool
           }
         }
       }
+      if (msgTools.length > 0) toolSequence.push(msgTools)
     }
   } catch { /* best-effort */ }
 
-  return { tools, bashCommands }
+  return { tools, bashCommands, toolSequence }
 }
 
 function getFirstUserMessage(db: SqliteDatabase, sessionId: string): string {
@@ -179,7 +183,7 @@ function createParser(source: SessionSource, seenKeys: Set<string>): SessionPars
         const model = config.model_name ?? 'unknown'
         const costUSD = calculateCost(model, inputTokens, outputTokens, 0, 0, 0)
 
-        const { tools, bashCommands } = extractToolsFromMessages(db, sessionId)
+        const { tools, bashCommands, toolSequence } = extractToolsFromMessages(db, sessionId)
         const userMessage = getFirstUserMessage(db, sessionId)
 
         const raw = session.updated_at || session.created_at || ''
@@ -200,6 +204,7 @@ function createParser(source: SessionSource, seenKeys: Set<string>): SessionPars
           costUSD,
           tools,
           bashCommands,
+          toolSequence: toolSequence.length > 1 ? toolSequence : undefined,
           timestamp: ts.toISOString(),
           speed: 'standard',
           deduplicationKey: dedupKey,

src/providers/kiro.ts

@@ -86,14 +86,17 @@ function parseChatFile(data: KiroChatFile, sessionId: string, project: string, s
 
   let pendingUserMessage = ''
   const allTools: string[] = []
+  const toolSequence: string[][] = []
 
   for (const msg of chat) {
     if (msg.role === 'human') {
       if (msg.content.startsWith('<identity>')) continue
       pendingUserMessage = msg.content.slice(0, 500)
     }
     if (msg.role === 'bot') {
-      allTools.push(...extractToolNames(msg.content))
+      const msgTools = extractToolNames(msg.content)
+      allTools.push(...msgTools)
+      if (msgTools.length > 0) toolSequence.push(msgTools)
     }
   }
 
@@ -125,6 +128,7 @@ function parseChatFile(data: KiroChatFile, sessionId: string, project: string, s
     costUSD,
     tools: [...new Set(allTools)],
     bashCommands: [],
+    toolSequence: toolSequence.length > 1 ? toolSequence : undefined,
     timestamp,
     speed: 'standard',
     deduplicationKey: dedupKey,

src/providers/mistral-vibe.ts

@@ -216,18 +216,21 @@ function parseToolArguments(raw: string | Record<string, unknown> | null | undef
   }
 }
 
-function extractTools(messages: VibeMessage[]): { tools: string[]; bashCommands: string[] } {
+function extractTools(messages: VibeMessage[]): { tools: string[]; bashCommands: string[]; toolSequence: string[][] } {
   const tools: string[] = []
   const bashCommands: string[] = []
+  const toolSequence: string[][] = []
 
   for (const message of messages) {
     if (message.role !== 'assistant') continue
+    const msgTools: string[] = []
     for (const toolCall of message.tool_calls ?? []) {
       const rawName = toolCall.function?.name
       if (!rawName) continue
 
       const mappedName = toolNameMap[rawName] ?? rawName
       tools.push(mappedName)
+      msgTools.push(mappedName)
 
       if (mappedName !== 'Bash') continue
       const args = parseToolArguments(toolCall.function?.arguments)
@@ -236,11 +239,13 @@ function extractTools(messages: VibeMessage[]): { tools: string[]; bashCommands:
         bashCommands.push(...extractBashCommands(command))
       }
     }
+    if (msgTools.length > 0) toolSequence.push(msgTools)
   }
 
   return {
     tools: [...new Set(tools)],
     bashCommands: [...new Set(bashCommands)],
+    toolSequence,
   }
 }
 
@@ -287,7 +292,7 @@ function createParser(source: SessionSource, seenKeys: Set<string>): SessionPars
 
       const messages = await readMessages(messagesPath)
       const model = resolveModel(metadata)
-      const { tools, bashCommands } = extractTools(messages)
+      const { tools, bashCommands, toolSequence } = extractTools(messages)
       const costUSD = calculateSessionCost(metadata, model, inputTokens, outputTokens)
 
       yield {
@@ -303,6 +308,7 @@ function createParser(source: SessionSource, seenKeys: Set<string>): SessionPars
         costUSD,
         tools,
         bashCommands,
+        toolSequence: toolSequence.length > 1 ? toolSequence : undefined,
         timestamp: metadata.end_time ?? metadata.start_time ?? '',
         speed: 'standard',
         deduplicationKey,

src/providers/types.ts

@@ -29,6 +29,7 @@ export type ParsedProviderCall = {
   sessionId: string
   project?: string
   projectPath?: string
+  toolSequence?: string[][]
 }
 
 export type Provider = {

src/session-cache.ts

@@ -29,6 +29,7 @@ export type CachedCall = {
   deduplicationKey: string
   project?: string
   projectPath?: string
+  toolSequence?: string[][]
 }
 
 export type CachedTurn = {

src/types.ts

@@ -84,6 +84,7 @@ export type ParsedApiCall = {
   bashCommands: string[]
   deduplicationKey: string
   cacheCreationOneHourTokens?: number
+  toolSequence?: string[][]
 }
 
 export type TaskCategory =

tests/classifier.test.ts

@@ -151,3 +151,46 @@ describe('classifyTurn — feature vs debugging precedence (#196)', () => {
     expect(c.category).toBe('debugging')
   })
 })
+
+describe('classifyTurn — retry detection via toolSequence', () => {
+  it('detects retries from multi-call turns (Claude-style)', () => {
+    const turn = makeTurn([
+      makeCall({ tools: ['Edit'] }),
+      makeCall({ tools: ['Bash'] }),
+      makeCall({ tools: ['Edit'] }),
+    ], 'fix the build')
+    const c = classifyTurn(turn)
+    expect(c.retries).toBe(1)
+  })
+
+  it('detects retries from toolSequence on a single call (Gemini/Mistral-style)', () => {
+    const call = makeCall({ tools: ['Edit', 'Bash'] })
+    call.toolSequence = [['Edit'], ['Bash'], ['Edit']]
+    const turn = makeTurn([call], 'fix the build')
+    const c = classifyTurn(turn)
+    expect(c.retries).toBe(1)
+  })
+
+  it('returns 0 retries for single call without toolSequence (aggregated tools)', () => {
+    const call = makeCall({ tools: ['Edit', 'Bash'] })
+    const turn = makeTurn([call], 'fix the build')
+    const c = classifyTurn(turn)
+    expect(c.retries).toBe(0)
+  })
+
+  it('counts multiple retries from toolSequence', () => {
+    const call = makeCall({ tools: ['Edit', 'Bash'] })
+    call.toolSequence = [['Edit'], ['Bash'], ['Edit'], ['Bash'], ['Edit']]
+    const turn = makeTurn([call], 'fix the build')
+    const c = classifyTurn(turn)
+    expect(c.retries).toBe(2)
+  })
+
+  it('ignores toolSequence with only one step', () => {
+    const call = makeCall({ tools: ['Edit', 'Bash'] })
+    call.toolSequence = [['Edit', 'Bash']]
+    const turn = makeTurn([call], 'fix the build')
+    const c = classifyTurn(turn)
+    expect(c.retries).toBe(0)
+  })
+})