ummaya 0.2.2 → 0.2.4

package/tui/src/services/api/claude.ts

@@ -128,6 +128,7 @@ import { feature } from 'bun:bundle'
 // aliases (ClientOptions, APIError, APIConnectionTimeoutError, APIUserAbortError
 // all re-exported by sdk-compat.ts as structural stubs).
 import type { ClientOptions } from '../../sdk-compat.js'
+import type { ReasoningMode } from '../../utils/kExaoneReasoning.js'
 import {
   APIConnectionTimeoutError,
   APIError,
@@ -216,9 +217,11 @@ import {
   TOOL_SEARCH_TOOL_NAME,
 } from '../../tools/ToolSearchTool/prompt.js'
 import {
-  isRootPrimitiveToolName,
+  getAdapterToolByName,
   selectTopKAdapterToolNamesForQuery,
 } from '../../tools/AdapterTool/AdapterTool.js'
+import { isNonSyntheticUserText } from '../../tools/_shared/citizenUserText.js'
+import { shouldSuppressUmmayaToolCallsForAnswerSynthesis } from '../../tools/_shared/toolChoiceRepair.js'
 import { count } from '../../utils/array.js'
 import { insertBlockAfterToolResults } from '../../utils/contentArray.js'
 import { validateBoundedIntEnvVar } from '../../utils/envValidation.js'
@@ -723,6 +726,7 @@ export type Options = {
   skipCacheWrite?: boolean
   temperatureOverride?: number
   effortValue?: EffortValue
+  reasoningMode?: ReasoningMode
   mcpTools: Tools
   hasPendingMcpServers?: boolean
   queryTracking?: QueryChainTracking
@@ -833,7 +837,7 @@ function latestUserTextForToolRetrieval(messages: Message[]): string {
     if (message?.type !== 'user') continue
     const content = message.message?.content
     if (typeof content === 'string') {
-      if (content.trim().length > 0) return content
+      if (isNonSyntheticUserText(content)) return content
       continue
     }
     if (Array.isArray(content)) {
@@ -844,7 +848,7 @@ function latestUserTextForToolRetrieval(messages: Message[]): string {
         )
         .map(block => block.text)
         .join('')
-      if (text.trim().length > 0) return text
+      if (isNonSyntheticUserText(text)) return text
     }
   }
   return ''
@@ -1183,10 +1187,36 @@ async function* queryModel(
     'query',
   )
 
-  // Precompute once — isDeferredTool does 2 GrowthBook lookups per call
+  const turnLocalAdapterToolNames = new Set(
+    selectTopKAdapterToolNamesForQuery(
+      latestUserTextForToolRetrieval(messages),
+    ),
+  )
+  if (options.toolChoice?.type === 'tool') {
+    turnLocalAdapterToolNames.add(options.toolChoice.name)
+  }
+  if (turnLocalAdapterToolNames.size > 0) {
+    logForDebugging(
+      `UMMAYA turn-local adapter schemas: ${[...turnLocalAdapterToolNames].join(', ')}`,
+    )
+  }
+  const requestTools =
+    turnLocalAdapterToolNames.size === 0
+      ? tools
+      : [
+          ...tools,
+          ...[...turnLocalAdapterToolNames]
+            .filter(toolName => !tools.some(tool => tool.name === toolName))
+            .map(toolName => getAdapterToolByName(toolName))
+            .filter((tool): tool is NonNullable<typeof tool> => Boolean(tool)),
+        ]
+
+  // Precompute once — isDeferredTool does 2 GrowthBook lookups per call.
+  // Include turn-local synced adapters even if the long-lived TUI tool pool
+  // was assembled before the latest backend manifest frame arrived.
   const deferredToolNames = new Set<string>()
   if (useToolSearch) {
-    for (const t of tools) {
+    for (const t of requestTools) {
       if (isDeferredTool(t)) deferredToolNames.add(t.name)
     }
   }
@@ -1204,32 +1234,29 @@ async function* queryModel(
     )
     useToolSearch = false
   }
-
-  const turnLocalAdapterToolNames = new Set(
-    selectTopKAdapterToolNamesForQuery(
-      latestUserTextForToolRetrieval(messages),
-    ),
-  )
-  if (turnLocalAdapterToolNames.size > 0) {
-    logForDebugging(
-      `UMMAYA turn-local adapter schemas: ${[...turnLocalAdapterToolNames].join(', ')}`,
-    )
+  const suppressUmmayaToolCalls =
+    shouldSuppressUmmayaToolCallsForAnswerSynthesis({ messages, tools: requestTools })
+  if (suppressUmmayaToolCalls) {
+    logForDebugging('UMMAYA suppressing tool schemas for answer synthesis')
   }
 
   // Filter out ToolSearchTool if tool search is not enabled for this model
   // ToolSearchTool returns tool_reference blocks which unsupported models can't handle
   let filteredTools: Tools
 
-  if (useToolSearch) {
+  if (suppressUmmayaToolCalls) {
+    filteredTools = []
+  } else if (useToolSearch) {
     // Dynamic tool loading: Only include deferred tools that have been discovered
     // via tool_reference blocks in the message history. This eliminates the need
     // to predeclare all deferred tools upfront and removes limits on tool quantity.
     const discoveredToolNames = extractDiscoveredToolNames(messages)
 
-    filteredTools = tools.filter(tool => {
-      if (turnLocalAdapterToolNames.size > 0 && isRootPrimitiveToolName(tool.name)) {
-        return false
-      }
+    filteredTools = requestTools.filter(tool => {
+      // 0.2.1 exposed the lightweight root primitives together with concrete
+      // adapter schemas. Keep that surface so K-EXAONE preserves CC-style
+      // prose→tool→prose loop painting, while still limiting concrete adapter
+      // schemas to the turn-local top-k set.
       if (turnLocalAdapterToolNames.has(tool.name)) return true
       // Always include non-deferred tools
       if (!deferredToolNames.has(tool.name)) return true
@@ -1239,11 +1266,10 @@ async function* queryModel(
       return discoveredToolNames.has(tool.name)
     })
   } else {
-    filteredTools = tools.filter(t => {
+    filteredTools = requestTools.filter(t => {
       if (toolMatchesName(t, TOOL_SEARCH_TOOL_NAME)) return false
-      if (turnLocalAdapterToolNames.size > 0 && isRootPrimitiveToolName(t.name)) {
-        return false
-      }
+      // Keep non-deferred root primitives even when concrete top-k adapter
+      // schemas are available; this matches the released 0.2.1 loop surface.
       if (isDeferredTool(t)) return turnLocalAdapterToolNames.has(t.name)
       return true
     })
@@ -1803,6 +1829,9 @@ async function* queryModel(
         output_config: outputConfig,
       }),
       ...(speed !== undefined && { speed }),
+      ...(options.reasoningMode !== undefined && {
+        reasoning_mode: options.reasoningMode,
+      }),
     }
   }
 
@@ -2346,7 +2375,7 @@ async function* queryModel(
                 max_tokens: maxOutputTokens,
               })
               yield createAssistantAPIErrorMessage({
-                content: `${API_ERROR_MESSAGE_PREFIX}: Claude's response exceeded the ${
+                content: `${API_ERROR_MESSAGE_PREFIX}: Ummaya's response exceeded the ${
                   maxOutputTokens
                 } output token maximum. To configure this behavior, set the CLAUDE_CODE_MAX_OUTPUT_TOKENS environment variable.`,
                 apiError: 'max_output_tokens',

package/tui/src/services/api/client.ts

@@ -26,6 +26,11 @@ import {
 } from '../../sdk-compat.js'
 import { assertFriendliApiKeyForUse } from '../../utils/auth.js'
 import { getUserAgent } from '../../utils/http.js'
+import {
+  providerReasoningPayload,
+  resolveKExaoneReasoningPolicy,
+  type ReasoningMode,
+} from '../../utils/kExaoneReasoning.js'
 import { UMMAYA_K_EXAONE_MODEL } from '../../utils/model/constants.js'
 
 export const CLIENT_REQUEST_ID_HEADER = 'x-client-request-id'
@@ -41,6 +46,11 @@ type RequestOptions = {
   headers?: Record<string, string>
 }
 
+type FriendliMessageStreamParams = BetaMessageStreamParams & {
+  stream?: boolean
+  reasoning_mode?: ReasoningMode
+}
+
 type FriendliClientArgs = {
   apiKey?: string
   maxRetries: number
@@ -130,7 +140,7 @@ class FriendliMessagesCompatClient {
   readonly beta: {
     messages: {
       create: (
-        params: BetaMessageStreamParams & { stream?: boolean },
+        params: FriendliMessageStreamParams,
         options?: RequestOptions,
       ) => unknown
     }
@@ -159,7 +169,7 @@ class FriendliMessagesCompatClient {
   }
 
   private async streamWithResponse(
-    params: BetaMessageStreamParams & { stream?: boolean },
+    params: FriendliMessageStreamParams,
     options?: RequestOptions,
   ): Promise<{
     data: Stream<BetaRawMessageStreamEvent>
@@ -188,7 +198,7 @@ class FriendliMessagesCompatClient {
   }
 
   private async complete(
-    params: BetaMessageStreamParams,
+    params: FriendliMessageStreamParams,
     options?: RequestOptions,
   ): Promise<BetaMessage> {
     const response = await this.fetchChatCompletion(
@@ -279,6 +289,9 @@ class FriendliMessagesCompatClient {
     const toolStates = new Map<number, ToolStreamState>()
     let finalUsage: BetaUsage | undefined
     let stopReason: BetaMessage['stop_reason'] = 'end_turn'
+    const allowReasoning = resolveKExaoneReasoningPolicy({
+      explicitSessionMode: params.reasoning_mode,
+    }).includeReasoning
 
     const ensureMessageStart = function* () {
       if (messageStarted) return
@@ -301,6 +314,12 @@ class FriendliMessagesCompatClient {
     const ensureTextBlock = function* () {
       yield* ensureMessageStart()
       if (textBlockIndex === undefined) {
+        if (
+          thinkingBlockIndex !== undefined &&
+          openedBlocks.has(thinkingBlockIndex)
+        ) {
+          yield* closeNonToolBlocks()
+        }
         textBlockIndex = nextBlockIndex++
         openedBlocks.add(textBlockIndex)
         yield {
@@ -314,6 +333,9 @@ class FriendliMessagesCompatClient {
     const ensureThinkingBlock = function* () {
       yield* ensureMessageStart()
       if (thinkingBlockIndex === undefined) {
+        if (textBlockIndex !== undefined && openedBlocks.has(textBlockIndex)) {
+          yield* closeNonToolBlocks()
+        }
         thinkingBlockIndex = nextBlockIndex++
         openedBlocks.add(thinkingBlockIndex)
         yield {
@@ -385,7 +407,7 @@ class FriendliMessagesCompatClient {
 
         for (const choice of chunk.choices ?? []) {
           const delta = choice.delta ?? {}
-          if (delta.reasoning_content) {
+          if (delta.reasoning_content && allowReasoning) {
             yield* ensureThinkingBlock()
             yield {
               type: 'content_block_delta' as const,
@@ -454,17 +476,20 @@ class FriendliMessagesCompatClient {
 }
 
 function buildOpenAIPayload(
-  params: BetaMessageStreamParams & { stream?: boolean },
+  params: FriendliMessageStreamParams,
 ): Record<string, unknown> {
   const messages = convertMessages(params.messages, params.system)
   const tools = convertTools(params.tools)
+  const reasoning = providerReasoningPayload(
+    resolveKExaoneReasoningPolicy({
+      explicitSessionMode: params.reasoning_mode,
+    }),
+  )
   const payload: Record<string, unknown> = {
     model: params.model || UMMAYA_K_EXAONE_MODEL,
     messages,
     max_tokens: params.max_tokens,
-    chat_template_kwargs: {
-      enable_thinking: isTruthy(process.env.UMMAYA_K_EXAONE_THINKING),
-    },
+    ...reasoning,
     ...(params.temperature !== undefined ? { temperature: params.temperature } : {}),
     ...(tools.length > 0
       ? {
@@ -740,7 +765,3 @@ function resolveFetch(fetchOverride: unknown): FetchLike {
 function trimSlash(value: string): string {
   return value.replace(/\/+$/, '')
 }
-
-function isTruthy(value: string | undefined): boolean {
-  return value === '1' || value?.toLowerCase() === 'true' || value?.toLowerCase() === 'yes'
-}

package/tui/src/services/api/ummaya.ts

@@ -55,8 +55,11 @@ export async function* queryModelWithStreaming(params: {
   const persistThinking = process.env.UMMAYA_PERSIST_THINKING === '1'
   let accumulatedThinking = ''
   let messageStartEmitted = false
-  let textBlockStarted = false
+  let nextContentBlockIndex = 0
+  let textBlockIndex: number | null = null
   let textBlockStopped = false
+  let thinkingBlockIndex: number | null = null
+  let thinkingBlockStopped = false
   const pendingContentBlocks: Array<{
     type: 'tool_use'
     id: string
@@ -119,16 +122,17 @@ export async function* queryModelWithStreaming(params: {
         }
         messageStartEmitted = true
       }
-      if (!textBlockStarted) {
+      if (thinkingText.length > 0 && thinkingBlockIndex === null) {
         yield {
           type: 'stream_event' as const,
           event: {
             type: 'content_block_start' as const,
-            index: 0,
-            content_block: { type: 'text' as const, text: '' },
+            index: nextContentBlockIndex,
+            content_block: { type: 'thinking' as const, thinking: '' },
           },
         }
-        textBlockStarted = true
+        thinkingBlockIndex = nextContentBlockIndex
+        nextContentBlockIndex += 1
       }
 
       if (thinkingText.length > 0) {
@@ -139,7 +143,7 @@ export async function* queryModelWithStreaming(params: {
           type: 'stream_event' as const,
           event: {
             type: 'content_block_delta' as const,
-            index: 0,
+            index: thinkingBlockIndex ?? 0,
             delta: { type: 'thinking_delta' as const, thinking: thinkingText },
           },
         }
@@ -147,21 +151,50 @@ export async function* queryModelWithStreaming(params: {
 
       accumulated += deltaText
       if (deltaText.length > 0) {
+        if (thinkingBlockIndex !== null && !thinkingBlockStopped) {
+          yield {
+            type: 'stream_event' as const,
+            event: {
+              type: 'content_block_stop' as const,
+              index: thinkingBlockIndex,
+            },
+          }
+          thinkingBlockStopped = true
+        }
+        if (textBlockIndex === null) {
+          yield {
+            type: 'stream_event' as const,
+            event: {
+              type: 'content_block_start' as const,
+              index: nextContentBlockIndex,
+              content_block: { type: 'text' as const, text: '' },
+            },
+          }
+          textBlockIndex = nextContentBlockIndex
+          nextContentBlockIndex += 1
+        }
         yield {
           type: 'stream_event' as const,
           event: {
             type: 'content_block_delta' as const,
-            index: 0,
+            index: textBlockIndex,
             delta: { type: 'text_delta' as const, text: deltaText },
           },
         }
       }
 
       if (frameAny.done) {
-        if (textBlockStarted && !textBlockStopped) {
+        if (thinkingBlockIndex !== null && !thinkingBlockStopped) {
+          yield {
+            type: 'stream_event' as const,
+            event: { type: 'content_block_stop' as const, index: thinkingBlockIndex },
+          }
+          thinkingBlockStopped = true
+        }
+        if (textBlockIndex !== null && !textBlockStopped) {
           yield {
             type: 'stream_event' as const,
-            event: { type: 'content_block_stop' as const, index: 0 },
+            event: { type: 'content_block_stop' as const, index: textBlockIndex },
           }
           textBlockStopped = true
         }
@@ -217,10 +250,17 @@ export async function* queryModelWithStreaming(params: {
         }
         messageStartEmitted = true
       }
-      if (textBlockStarted && !textBlockStopped) {
+      if (thinkingBlockIndex !== null && !thinkingBlockStopped) {
         yield {
           type: 'stream_event' as const,
-          event: { type: 'content_block_stop' as const, index: 0 },
+          event: { type: 'content_block_stop' as const, index: thinkingBlockIndex },
+        }
+        thinkingBlockStopped = true
+      }
+      if (textBlockIndex !== null && !textBlockStopped) {
+        yield {
+          type: 'stream_event' as const,
+          event: { type: 'content_block_stop' as const, index: textBlockIndex },
         }
         textBlockStopped = true
       }
@@ -232,7 +272,8 @@ export async function* queryModelWithStreaming(params: {
       }
 
       pendingContentBlocks.push(toolUseBlock)
-      const toolBlockIndex = textBlockStarted ? 1 : 0
+      const toolBlockIndex = nextContentBlockIndex
+      nextContentBlockIndex += 1
 
       yield {
         type: 'stream_event' as const,
@@ -271,10 +312,17 @@ export async function* queryModelWithStreaming(params: {
     } else if (frameAny.kind === 'error') {
       const reason = frameAny.message ?? 'UMMAYA backend error'
       yield createAssistantMessage({ content: `[UMMAYA backend error] ${reason}` })
-      if (textBlockStarted && !textBlockStopped) {
+      if (thinkingBlockIndex !== null && !thinkingBlockStopped) {
         yield {
           type: 'stream_event' as const,
-          event: { type: 'content_block_stop' as const, index: 0 },
+          event: { type: 'content_block_stop' as const, index: thinkingBlockIndex },
+        }
+        thinkingBlockStopped = true
+      }
+      if (textBlockIndex !== null && !textBlockStopped) {
+        yield {
+          type: 'stream_event' as const,
+          event: { type: 'content_block_stop' as const, index: textBlockIndex },
         }
         textBlockStopped = true
       }
@@ -299,10 +347,16 @@ export async function* queryModelWithStreaming(params: {
     })
   }
   if (messageStartEmitted) {
-    if (textBlockStarted && !textBlockStopped) {
+    if (thinkingBlockIndex !== null && !thinkingBlockStopped) {
+      yield {
+        type: 'stream_event' as const,
+        event: { type: 'content_block_stop' as const, index: thinkingBlockIndex },
+      }
+    }
+    if (textBlockIndex !== null && !textBlockStopped) {
       yield {
         type: 'stream_event' as const,
-        event: { type: 'content_block_stop' as const, index: 0 },
+        event: { type: 'content_block_stop' as const, index: textBlockIndex },
       }
     }
     yield {

package/tui/src/skills/bundled/stuck.ts

@@ -1,15 +1,15 @@
 import { registerBundledSkill } from '../bundledSkills.js'
 
-// Prompt text contains `ps` commands as instructions for Claude to run,
+// Prompt text contains `ps` commands as instructions for UMMAYA to run,
 // not commands this file executes.
 // eslint-disable-next-line custom-rules/no-direct-ps-commands
-const STUCK_PROMPT = `# /stuck — diagnose frozen/slow Claude Code sessions
+const STUCK_PROMPT = `# /stuck — diagnose frozen/slow UMMAYA sessions
 
-The user thinks another Claude Code session on this machine is frozen, stuck, or very slow. Investigate and post a report to #claude-code-feedback.
+The user thinks another UMMAYA session on this machine is frozen, stuck, or very slow. Investigate and post a report to the configured UMMAYA feedback channel.
 
 ## What to look for
 
-Scan for other Claude Code processes (excluding the current one — PID is in \`process.pid\` but for shell commands just exclude the PID you see running this prompt). Process names are typically \`claude\` (installed) or \`cli\` (native dev build).
+Scan for other UMMAYA processes (excluding the current one — PID is in \`process.pid\` but for shell commands just exclude the PID you see running this prompt). Process names are typically \`ummaya\` (installed) or \`cli\` (native dev build).
 
 Signs of a stuck session:
 - **High CPU (≥90%) sustained** — likely an infinite loop. Sample twice, 1-2s apart, to confirm it's not a transient spike.
@@ -21,17 +21,17 @@ Signs of a stuck session:
 
 ## Investigation steps
 
-1. **List all Claude Code processes** (macOS/Linux):
+1. **List all UMMAYA processes** (macOS/Linux):
    \`\`\`
-   ps -axo pid=,pcpu=,rss=,etime=,state=,comm=,command= | grep -E '(claude|cli)' | grep -v grep
+   ps -axo pid=,pcpu=,rss=,etime=,state=,comm=,command= | grep -E '(ummaya|cli)' | grep -v grep
    \`\`\`
-   Filter to rows where \`comm\` is \`claude\` or (\`cli\` AND the command path contains "claude").
+   Filter to rows where \`comm\` is \`ummaya\` or (\`cli\` AND the command path contains "ummaya").
 
 2. **For anything suspicious**, gather more context:
    - Child processes: \`pgrep -lP <pid>\`
    - If high CPU: sample again after 1-2s to confirm it's sustained
    - If a child looks hung (e.g., a git command), note its full command line with \`ps -p <child_pid> -o command=\`
-   - Check the session's debug log if you can infer the session ID: \`~/.claude/debug/<session-id>.txt\` (the last few hundred lines often show what it was doing before hanging)
+   - Check the session's debug log if you can infer the session ID: \`~/.ummaya/debug/<session-id>.txt\` (the last few hundred lines often show what it was doing before hanging)
 
 3. **Consider a stack dump** for a truly frozen process (advanced, optional):
    - macOS: \`sample <pid> 3\` gives a 3-second native stack sample
@@ -41,17 +41,17 @@ Signs of a stuck session:
 
 **Only post to Slack if you actually found something stuck.** If every session looks healthy, tell the user that directly — do not post an all-clear to the channel.
 
-If you did find a stuck/slow session, post to **#claude-code-feedback** (channel ID: \`C07VBSHV7EV\`) using the Slack MCP tool. Use ToolSearch to find \`slack_send_message\` if it's not already loaded.
+If you did find a stuck/slow session, post to the configured UMMAYA feedback channel using the Slack MCP tool. Use ToolSearch to find \`slack_send_message\` if it's not already loaded.
 
 **Use a two-message structure** to keep the channel scannable:
 
-1. **Top-level message** — one short line: hostname, Claude Code version, and a terse symptom (e.g. "session PID 12345 pegged at 100% CPU for 10min" or "git subprocess hung in D state"). No code blocks, no details.
+1. **Top-level message** — one short line: hostname, UMMAYA version, and a terse symptom (e.g. "session PID 12345 pegged at 100% CPU for 10min" or "git subprocess hung in D state"). No code blocks, no details.
 2. **Thread reply** — the full diagnostic dump. Pass the top-level message's \`ts\` as \`thread_ts\`. Include:
    - PID, CPU%, RSS, state, uptime, command line, child processes
    - Your diagnosis of what's likely wrong
    - Relevant debug log tail or \`sample\` output if you captured it
 
-If Slack MCP isn't available, format the report as a message the user can copy-paste into #claude-code-feedback (and let them know to thread the details themselves).
+If Slack MCP isn't available, format the report as a message the user can copy-paste into the configured UMMAYA feedback channel (and let them know to thread the details themselves).
 
 ## Notes
 - Don't kill or signal any processes — this is diagnostic only.
@@ -66,7 +66,7 @@ export function registerStuckSkill(): void {
   registerBundledSkill({
     name: 'stuck',
     description:
-      '[ANT-ONLY] Investigate frozen/stuck/slow Claude Code sessions on this machine and post a diagnostic report to #claude-code-feedback.',
+      '[ANT-ONLY] Investigate frozen/stuck/slow UMMAYA sessions on this machine and post a diagnostic report to the configured UMMAYA feedback channel.',
     userInvocable: true,
     async getPromptForCommand(args) {
       let prompt = STUCK_PROMPT

package/tui/src/state/AppStateStore.ts

@@ -35,6 +35,10 @@ import type { DenialTrackingState } from '../utils/permissions/denialTracking.js
 import type { PermissionMode } from '../utils/permissions/PermissionMode.js'
 import { getInitialSettings } from '../utils/settings/settings.js'
 import type { SettingsJson } from '../utils/settings/types.js'
+import {
+  getInitialReasoningModeSetting,
+  type ReasoningMode,
+} from '../utils/kExaoneReasoning.js'
 import { shouldEnableThinkingByDefault } from '../utils/thinking.js'
 import type { Store } from './store.js'
 
@@ -425,6 +429,8 @@ export type AppState = DeepImmutable<{
   advisorModel?: string
   // Effort value
   effortValue?: EffortValue
+  // K-EXAONE/FriendliAI reasoning payload policy.
+  reasoningMode?: ReasoningMode
   // Set synchronously in launchUltraplan before the detached flow starts.
   // Prevents duplicate launches during the ~5s window before
   // ultraplanSessionUrl is set by teleportToRemote. Cleared by launchDetached
@@ -563,6 +569,7 @@ export function getDefaultAppState(): AppState {
     authVersion: 0,
     initialMessage: null,
     effortValue: undefined,
+    reasoningMode: getInitialReasoningModeSetting(),
     activeOverlays: new Set<string>(),
     fastMode: false,
   }