ummaya 0.2.4 → 0.2.6

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

@@ -1,418 +1,26 @@
-import { randomUUID } from 'crypto'
-import { APIUserAbortError } from 'src/sdk-compat.js'
-import {
-  getOrCreateUmmayaBridge,
-  getUmmayaBridgeSessionId,
-} from '../../ipc/bridgeSingleton.js'
-import type {
-  ChatRequestFrame,
-  IPCFrame,
-} from '../../ipc/frames.generated.js'
-import { getToolDefinitionsForFrame } from '../../query/toolSerialization.js'
-import { SYNTHETIC_MODEL } from '../../utils/messageText.js'
-import { createAssistantMessage } from '../../utils/assistantMessageFactories.js'
-import { buildChatMessagesFromTranscript } from '../../query/chatMessagesBuilder.js'
-import { assertFriendliApiKeyForUse } from '../../utils/auth.js'
-
-export async function* queryModelWithStreaming(params: {
-  messages: readonly unknown[]
-  systemPrompt: unknown
-  thinkingConfig?: unknown
-  tools?: unknown
-  signal?: AbortSignal
-  options?: { model?: string; querySource?: string; [k: string]: unknown }
-}): AsyncGenerator<unknown> {
-  const { messages, signal } = params
-  assertFriendliApiKeyForUse()
-
-  const correlationId = randomUUID()
-  const messageUuid = randomUUID()
-  const innerMessageId = randomUUID()
-  const turnStartedAt = performance.now()
-  const chatMessages = buildChatMessagesFromTranscript(messages)
-  const bridge = getOrCreateUmmayaBridge()
-  const sessionId = getUmmayaBridgeSessionId()
-  const tools = await getToolDefinitionsForFrame()
-
-  const frame: ChatRequestFrame = {
-    session_id: sessionId,
-    correlation_id: correlationId,
-    ts: new Date().toISOString(),
-    role: 'tui',
-    kind: 'chat_request',
-    messages: chatMessages as ChatRequestFrame['messages'],
-    tools: tools as ChatRequestFrame['tools'],
-  }
-
-  yield { type: 'stream_request_start' as const }
-
-  const sent = bridge.send(frame as unknown as IPCFrame)
-  if (!sent) {
-    throw new Error('UMMAYA bridge send failed (backend exited)')
-  }
-
-  let accumulated = ''
-  const persistThinking = process.env.UMMAYA_PERSIST_THINKING === '1'
-  let accumulatedThinking = ''
-  let messageStartEmitted = 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
-    name: string
-    input: unknown
-  }> = []
-
-  const hasAssistantPayload = (): boolean =>
-    accumulated.trimStart().length > 0 ||
-    pendingContentBlocks.length > 0 ||
-    (persistThinking && accumulatedThinking.length > 0)
-
-  for await (const f of bridge.frames()) {
-    if (signal?.aborted) {
-      throw new APIUserAbortError()
-    }
-
-    const frameAny = f as {
-      kind?: string
-      correlation_id?: string
-      delta?: string
-      thinking?: string
-      done?: boolean
-      message?: string
-      call_id?: string
-      name?: string
-      arguments?: unknown
-      envelope?: { kind?: string; [k: string]: unknown }
-    }
-
-    if (frameAny.correlation_id !== correlationId) continue
-
-    if (frameAny.kind === 'assistant_chunk') {
-      const deltaText = frameAny.delta ?? ''
-      const thinkingText = frameAny.thinking ?? ''
-
-      if (!messageStartEmitted) {
-        const ttftMs = performance.now() - turnStartedAt
-        yield {
-          type: 'stream_event' as const,
-          event: {
-            type: 'message_start' as const,
-            message: {
-              id: innerMessageId,
-              type: 'message',
-              role: 'assistant',
-              content: [],
-              model: SYNTHETIC_MODEL,
-              stop_reason: null,
-              stop_sequence: null,
-              usage: {
-                input_tokens: 0,
-                output_tokens: 0,
-                cache_creation_input_tokens: 0,
-                cache_read_input_tokens: 0,
-              },
-            },
-          },
-          ttftMs,
-        }
-        messageStartEmitted = true
-      }
-      if (thinkingText.length > 0 && thinkingBlockIndex === null) {
-        yield {
-          type: 'stream_event' as const,
-          event: {
-            type: 'content_block_start' as const,
-            index: nextContentBlockIndex,
-            content_block: { type: 'thinking' as const, thinking: '' },
-          },
-        }
-        thinkingBlockIndex = nextContentBlockIndex
-        nextContentBlockIndex += 1
-      }
-
-      if (thinkingText.length > 0) {
-        if (persistThinking) {
-          accumulatedThinking += thinkingText
-        }
-        yield {
-          type: 'stream_event' as const,
-          event: {
-            type: 'content_block_delta' as const,
-            index: thinkingBlockIndex ?? 0,
-            delta: { type: 'thinking_delta' as const, thinking: thinkingText },
-          },
-        }
-      }
-
-      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: textBlockIndex,
-            delta: { type: 'text_delta' as const, text: deltaText },
-          },
-        }
-      }
-
-      if (frameAny.done) {
-        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: textBlockIndex },
-          }
-          textBlockStopped = true
-        }
-        if (hasAssistantPayload() || !messageStartEmitted) {
-          const finalMessage = createFinalAssistantMessage({
-            accumulated,
-            accumulatedThinking,
-            messageUuid,
-            innerMessageId,
-            pendingContentBlocks,
-            persistThinking,
-          })
-          yield finalMessage
-        }
-        yield {
-          type: 'stream_event' as const,
-          event: {
-            type: 'message_delta' as const,
-            delta: { stop_reason: 'end_turn', stop_sequence: null },
-            usage: { output_tokens: 0 },
-          },
-        }
-        yield {
-          type: 'stream_event' as const,
-          event: { type: 'message_stop' as const },
-        }
-        return
-      }
-    } else if (frameAny.kind === 'tool_call') {
-      if (!messageStartEmitted) {
-        const ttftMs = performance.now() - turnStartedAt
-        yield {
-          type: 'stream_event' as const,
-          event: {
-            type: 'message_start' as const,
-            message: {
-              id: innerMessageId,
-              type: 'message',
-              role: 'assistant',
-              content: [],
-              model: SYNTHETIC_MODEL,
-              stop_reason: null,
-              stop_sequence: null,
-              usage: {
-                input_tokens: 0,
-                output_tokens: 0,
-                cache_creation_input_tokens: 0,
-                cache_read_input_tokens: 0,
-              },
-            },
-          },
-          ttftMs,
-        }
-        messageStartEmitted = true
-      }
-      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: textBlockIndex },
-        }
-        textBlockStopped = true
-      }
-      const toolUseBlock = {
-        type: 'tool_use' as const,
-        id: frameAny.call_id ?? '',
-        name: frameAny.name ?? '(unknown tool)',
-        input: frameAny.arguments ?? {},
-      }
-
-      pendingContentBlocks.push(toolUseBlock)
-      const toolBlockIndex = nextContentBlockIndex
-      nextContentBlockIndex += 1
-
-      yield {
-        type: 'stream_event' as const,
-        event: {
-          type: 'content_block_start' as const,
-          index: toolBlockIndex,
-          content_block: toolUseBlock,
-        },
-      }
-      yield {
-        type: 'stream_event' as const,
-        event: { type: 'content_block_stop' as const, index: toolBlockIndex },
-      }
-
-      yield createFinalAssistantMessage({
-        accumulated,
-        accumulatedThinking,
-        messageUuid,
-        innerMessageId,
-        pendingContentBlocks,
-        persistThinking,
-      })
-      yield {
-        type: 'stream_event' as const,
-        event: {
-          type: 'message_delta' as const,
-          delta: { stop_reason: 'tool_use', stop_sequence: null },
-          usage: { output_tokens: 0 },
-        },
-      }
-      yield {
-        type: 'stream_event' as const,
-        event: { type: 'message_stop' as const },
-      }
-      return
-    } else if (frameAny.kind === 'error') {
-      const reason = frameAny.message ?? 'UMMAYA backend error'
-      yield createAssistantMessage({ content: `[UMMAYA backend error] ${reason}` })
-      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: textBlockIndex },
-        }
-        textBlockStopped = true
-      }
-      if (messageStartEmitted) {
-        yield {
-          type: 'stream_event' as const,
-          event: { type: 'message_stop' as const },
-        }
-      }
-      return
-    }
-  }
-
-  if (hasAssistantPayload() || !messageStartEmitted) {
-    yield createFinalAssistantMessage({
-      accumulated,
-      accumulatedThinking,
-      messageUuid,
-      innerMessageId,
-      pendingContentBlocks,
-      persistThinking,
-    })
-  }
-  if (messageStartEmitted) {
-    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: textBlockIndex },
-      }
-    }
-    yield {
-      type: 'stream_event' as const,
-      event: { type: 'message_stop' as const },
-    }
-  }
-}
-
-function createFinalAssistantMessage({
-  accumulated,
-  accumulatedThinking,
-  messageUuid,
-  innerMessageId,
-  pendingContentBlocks,
-  persistThinking,
-}: {
-  accumulated: string
-  accumulatedThinking: string
-  messageUuid: string
-  innerMessageId: string
-  pendingContentBlocks: Array<{
-    type: 'tool_use'
-    id: string
-    name: string
-    input: unknown
-  }>
-  persistThinking: boolean
-}): unknown {
-  const trimmedText = accumulated.trimStart()
-  type AssistantBlock =
-    | { type: 'thinking'; thinking: string }
-    | { type: 'text'; text: string }
-    | { type: 'tool_use'; id: string; name: string; input: unknown }
-
-  const blocks: AssistantBlock[] = []
-  if (persistThinking && accumulatedThinking.length > 0) {
-    blocks.push({ type: 'thinking', thinking: accumulatedThinking })
-  }
-  if (trimmedText.length > 0) {
-    blocks.push({ type: 'text', text: trimmedText })
-  }
-  for (const toolUse of pendingContentBlocks) {
-    blocks.push(toolUse)
-  }
-
-  const finalContent =
-    blocks.length > 0
-      ? (blocks as Parameters<typeof createAssistantMessage>[0]['content'])
-      : trimmedText
-
-  const finalMessage = createAssistantMessage({ content: finalContent }) as {
-    uuid: string
-    message: { id: string }
-  }
-  finalMessage.uuid = messageUuid
-  finalMessage.message.id = innerMessageId
-  return finalMessage
-}
+export {
+  addCacheBreakpoints,
+  assistantMessageToMessageParam,
+  buildSystemPromptBlocks,
+  getCacheControl,
+  getPromptCachingEnabled,
+  stripExcessMediaItems,
+  userMessageToMessageParam,
+} from './ummaya/messages.js'
+export {
+  adjustParamsForNonStreaming,
+  executeNonStreamingRequest,
+  getMaxOutputTokensForModel,
+  MAX_NON_STREAMING_TOKENS,
+  queryHaiku,
+  queryModelWithoutStreaming,
+  queryWithModel,
+} from './ummaya/nonStreaming.js'
+export {
+  configureTaskBudgetParams,
+  getAPIMetadata,
+  getExtraBodyParams,
+  queryModelWithStreaming,
+  verifyApiKey,
+} from './ummaya/provider.js'
+export { accumulateUsage, cleanupStream, updateUsage } from './ummaya/usage.js'

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

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: Apache-2.0
 // UMMAYA keeps Claude Code's query/provider boundary: query.ts owns the
 // agentic loop and this helper preserves the SDK-shaped retry generator used
-// by services/api/claude.ts. Provider-specific retry policy can be widened
+// by services/api/ummaya.ts. Provider-specific retry policy can be widened
 // later, but the control-flow shape must stay generator-based so query.ts sees
 // the same stream/system-message contract as the CC restored source.
 

package/tui/src/services/awaySummary.ts

@@ -2,7 +2,7 @@
 // Research use — adapted from Claude Code 2.1.88 src/services/awaySummary.ts
 //
 // UMMAYA Epic #2077 — awaySummary restored from CC source. Now that
-// services/api/claude.ts is restored with queryModelWithoutStreaming routing
+// services/api/ummaya.ts is restored with queryModelWithoutStreaming routing
 // through the stdio bridge, this file can call it directly as in CC 2.1.88.
 // Prior stub (Epic #1633) returned null unconditionally.
 
@@ -14,7 +14,7 @@ import { getAssistantMessageText } from '../utils/messageContent.js'
 import { createUserMessage } from '../utils/userMessageFactories.js'
 import { getSmallFastModel } from '../utils/model/model.js'
 import { asSystemPrompt } from '../utils/systemPromptType.js'
-import { queryModelWithoutStreaming } from './api/claude.js'
+import { queryModelWithoutStreaming } from './api/ummaya.js'
 import { getSessionMemoryContent } from './SessionMemory/sessionMemoryUtils.js'
 
 // Recap only needs recent context — truncate to avoid "prompt too long" on

package/tui/src/services/claudeAiLimits.ts

@@ -10,7 +10,7 @@ import { getSmallFastModel } from '../utils/model/model.js'
 import { isEssentialTrafficOnly } from '../utils/privacyLevel.js'
 import type { AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS } from './analytics/index.js'
 import { logEvent } from './analytics/index.js'
-import { getAPIMetadata } from './api/claude.js'
+import { getAPIMetadata } from './api/ummaya.js'
 import { getAnthropicClient } from './api/client.js'
 import {
   processRateLimitHeaders,

package/tui/src/services/compact/autoCompact.ts

@@ -13,7 +13,7 @@ import type { CacheSafeParams } from '../../utils/forkedAgent.js'
 import { logError } from '../../utils/log.js'
 import { tokenCountWithEstimation } from '../../utils/tokens.js'
 import { getFeatureValue_CACHED_MAY_BE_STALE } from '../analytics/growthbook.js'
-import { getMaxOutputTokensForModel } from '../api/claude.js'
+import { getMaxOutputTokensForModel } from '../api/ummaya.js'
 import { notifyCompaction } from '../api/promptCacheBreakDetection.js'
 import { setLastSummarizedMessageId } from '../SessionMemory/sessionMemoryUtils.js'
 import {

package/tui/src/services/compact/compact.ts

@@ -99,7 +99,7 @@ import {
 import {
   getMaxOutputTokensForModel,
   queryModelWithStreaming,
-} from '../api/claude.js'
+} from '../api/ummaya.js'
 import {
   getPromptTooLongTokenGap,
   PROMPT_TOO_LONG_ERROR_MESSAGE,

package/tui/src/services/lsp/types.ts

@@ -1,36 +1,14 @@
 // [P0 reconstructed · rebuild-stubs.ts · symbol-complete stub]
 // Aggregated from every consumer import across src/.
-/* eslint-disable @typescript-eslint/no-explicit-any */
+import { createStub } from '../../stubs/any-stub'
 
-const __noop = (..._args: unknown[]): any => undefined as any;
-const __stub: any = new Proxy(function () {} as any, {
-  get(_t, p) {
-    if (p === Symbol.toPrimitive) return () => ""
-    if (p === Symbol.iterator) return function* () {}
-    if (p === Symbol.asyncIterator) return async function* () {}
-    if (p === Symbol.toStringTag) return "Stub"
-    if (p === Symbol.for("nodejs.util.inspect.custom")) return () => "<Stub>"
-    if (p === "inspect") return () => "<Stub>"
-    if (p === "then") return undefined
-    if (p === "toString") return () => ""
-    if (p === "valueOf") return () => undefined
-    if (p === "toJSON") return () => null
-    if (p === "length") return 0
-    if (p === "name") return "Stub"
-    if (p === "message") return ""
-    if (p === "stack") return ""
-    if (p === "constructor") return Object
-    return __stub
-  },
-  apply() { return __stub },
-  construct() { return __stub },
-});
+const __stub = createStub()
 
-export type LspServerConfig = any;
-export const LspServerConfig: any = __stub;
-export type LspServerState = any;
-export const LspServerState: any = __stub;
-export type ScopedLspServerConfig = any;
-export const ScopedLspServerConfig: any = __stub;
+export type LspServerConfig = unknown;
+export const LspServerConfig = __stub;
+export type LspServerState = unknown;
+export const LspServerState = __stub;
+export type ScopedLspServerConfig = unknown;
+export const ScopedLspServerConfig = __stub;
 
 export default __stub;

package/tui/src/services/tips/types.ts

@@ -1,34 +1,12 @@
 // [P0 reconstructed · rebuild-stubs.ts · symbol-complete stub]
 // Aggregated from every consumer import across src/.
-/* eslint-disable @typescript-eslint/no-explicit-any */
+import { createStub } from '../../stubs/any-stub'
 
-const __noop = (..._args: unknown[]): any => undefined as any;
-const __stub: any = new Proxy(function () {} as any, {
-  get(_t, p) {
-    if (p === Symbol.toPrimitive) return () => ""
-    if (p === Symbol.iterator) return function* () {}
-    if (p === Symbol.asyncIterator) return async function* () {}
-    if (p === Symbol.toStringTag) return "Stub"
-    if (p === Symbol.for("nodejs.util.inspect.custom")) return () => "<Stub>"
-    if (p === "inspect") return () => "<Stub>"
-    if (p === "then") return undefined
-    if (p === "toString") return () => ""
-    if (p === "valueOf") return () => undefined
-    if (p === "toJSON") return () => null
-    if (p === "length") return 0
-    if (p === "name") return "Stub"
-    if (p === "message") return ""
-    if (p === "stack") return ""
-    if (p === "constructor") return Object
-    return __stub
-  },
-  apply() { return __stub },
-  construct() { return __stub },
-});
+const __stub = createStub()
 
-export type Tip = any;
-export const Tip: any = __stub;
-export type TipContext = any;
-export const TipContext: any = __stub;
+export type Tip = unknown;
+export const Tip = __stub;
+export type TipContext = unknown;
+export const TipContext = __stub;
 
 export default __stub;

package/tui/src/services/tokenEstimation.ts

@@ -23,7 +23,7 @@ import {
 } from '../utils/model/model.js'
 import { jsonStringify } from '../utils/slowOperations.js'
 import { isToolReferenceBlock } from '../utils/toolSearch.js'
-import { getAPIMetadata, getExtraBodyParams } from './api/claude.js'
+import { getAPIMetadata, getExtraBodyParams } from './api/ummaya.js'
 import { getAnthropicClient } from './api/client.js'
 import { withTokenCountVCR } from './vcr.js'
 

package/tui/src/services/toolRegistry/bootGuard.ts

@@ -2,7 +2,7 @@
 // Epic γ #2294 · T004 · ToolRegistry boot guard.
 //
 // Walks every active registered UMMAYA primitive (find / locate /
-// send / check) at process boot and asserts that
+// send / check / document) at process boot and asserts that
 // each one exposes the full `Tool<>` 10-member surface from
 // `tui/src/Tool.ts` (byte-identical to CC `Tool.ts`). Fails closed with a
 // Korean diagnostic if any member is missing.
@@ -22,7 +22,7 @@
 
 import type { Tool } from '../../Tool.js'
 
-const PRIMITIVE_NAMES = ['find', 'locate', 'send', 'check'] as const
+const PRIMITIVE_NAMES = ['find', 'locate', 'send', 'check', 'document'] as const
 export type PrimitiveName = (typeof PRIMITIVE_NAMES)[number]
 
 const REQUIRED_MEMBERS = [
@@ -58,8 +58,8 @@ export type BootResult =
  * structured `BootResult`. Caller decides whether to `process.exit(1)` or
  * throw — the guard itself has no side effects beyond reading the registry.
  *
- * Performance: O(P × M) where P = number of active primitives (4) and M = required
- * member count (9) — bounded at 36 property reads. Wall-clock budget on a
+ * Performance: O(P × M) where P = number of active primitives (5) and M = required
+ * member count (10) — bounded at 50 property reads. Wall-clock budget on a
  * developer laptop: ≤ 200 ms (Spec SC-002).
  */
 export function verifyBootRegistry(registry: readonly Tool[]): BootResult {
@@ -82,7 +82,7 @@ export function verifyBootRegistry(registry: readonly Tool[]): BootResult {
       diagnostic:
         `[UMMAYA][bootGuard] 활성 primitive 중 일부가 ToolRegistry에 등록되지 않았습니다. ` +
         `누락: ${missingNames.join(', ')}.\n` +
-        `UMMAYA는 활성 primitive(find/locate/send/check) 모두 등록되어야 부팅을 허용합니다.\n` +
+        `UMMAYA는 활성 primitive(find/locate/send/check/document) 모두 등록되어야 부팅을 허용합니다.\n` +
         `참조: specs/2294-5-primitive-align/contracts/registry-boot-guard.md`,
     }
   }

package/tui/src/services/toolUseSummary/toolUseSummaryGenerator.ts

@@ -10,7 +10,7 @@ import { toError } from '../../utils/errors.js'
 import { logError } from '../../utils/log.js'
 import { jsonStringify } from '../../utils/slowOperations.js'
 import { asSystemPrompt } from '../../utils/systemPromptType.js'
-import { queryHaiku } from '../api/claude.js'
+import { queryHaiku } from '../api/ummaya.js'
 
 const TOOL_USE_SUMMARY_SYSTEM_PROMPT = `Write a short summary label describing what these tool calls accomplished. It appears as a single-line row in a mobile app and truncates around 30 characters, so think git-commit-subject, not sentence.