whale-code 6.5.5 → 6.5.7

package/dist/shared/api-client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"api-client.js","names":["getContextManagement","getMaxOutputTokens","getThinkingConfig","isRetryableError","addPromptCaching","getProvider","MAX_RETRIES","RETRY_BASE_DELAY_MS","buildAPIRequest","opts","model","contextProfile","thinkingEnabled","maxOutputTokens","contextOverrides","provider","isAnthropicModel","betas","edits","ctxMgmt","config","type","keep","value","trigger","supportsCompaction","includes","push","pause_after_compaction","instructions","thinkingCfg","beta","maxTokens","thinking","budget_tokens","Math","min","filter","e","contextManagement","callServerProxy","proxyUrl","signal","timeoutMs","tokenRefreshed","attempt","body","mode","messages","system","tools","max_tokens","apiConfig","stream","length","context_management","temperature","storeId","store_id","Array","isArray","msgs","keepFrom","max","i","msg","content","block","source","text","sub","serialized","JSON","stringify","fetchOpts","method","headers","token","String","Buffer","byteLength","controller","timeout","AbortController","setTimeout","abort","response","fetch","clearTimeout","ok","errorBody","status","onTokenRefresh","newToken","Object","assign","Error","err","aborted","errMsg","message","onRetry","delay","pow","Promise","resolve","reject","reason","timer","onAbort","addEventListener","once","fallbackModel","fromModel","onFallback","buildSystemBlocks","systemPrompt","costContext","enableCaching","cache_control","prepareWithCaching","trimGeminiContext","estimatedTokens","threshold","keepRecent","toolResultCount","trimCount","trimmed","map","callTranscribe","audioBase64","mediaType","Authorization","audio_base64","media_type","errText","data","json","success","transcript","error","trimOpenAIContext","COMPACTION_TIMEOUT_MS","requestProviderCompaction","system_prompt","compaction_content"],"sources":["../../src/shared/api-client.ts"],"sourcesContent":["/**\n * Unified API Client — ONE server proxy caller with retry and config builder.\n *\n * Replaces proxy+direct call paths in agent-loop, subagent, teammate.\n * CLI-only: always calls through server proxy (no direct Anthropic SDK usage).\n */\n\nimport {\n  getContextManagement,\n  getMaxOutputTokens,\n  getThinkingConfig,\n  isRetryableError,\n  addPromptCaching,\n} from \"./agent-core.js\";\nimport type { CompactableMessage } from \"./compaction.js\";\nimport { getProvider } from \"./constants.js\";\nimport type { ContextProfile, APIRequestConfig } from \"./types.js\";\n\n// ============================================================================\n// CONSTANTS\n// ============================================================================\n\nconst MAX_RETRIES = 3;\nconst RETRY_BASE_DELAY_MS = 1000;\n\n// ============================================================================\n// API REQUEST CONFIG BUILDER\n// ============================================================================\n\n/**\n * Build all Anthropic API config from a simple profile.\n *\n * Profiles:\n * - 'main': clear at 80K/keep 3, compact at 120K (Opus only)\n * - 'subagent': clear at 60K/keep 2, no compaction, 8192 max tokens\n * - 'teammate': clear at 80K/keep 3, no compaction\n */\nexport function buildAPIRequest(opts: {\n  model: string;\n  contextProfile: ContextProfile;\n  thinkingEnabled?: boolean;\n  maxOutputTokens?: number;\n  contextOverrides?: import(\"./agent-core.js\").ContextManagementOverrides;\n}): APIRequestConfig {\n  const { model, contextProfile, thinkingEnabled = false, maxOutputTokens, contextOverrides } = opts;\n\n  // Context management config per profile\n  // Non-Anthropic models (Gemini, OpenAI, etc.) don't support Anthropic betas or context management\n  const provider = getProvider(model);\n  const isAnthropicModel = provider === \"anthropic\" || provider === \"bedrock\";\n  let betas: string[] = isAnthropicModel ? [\"context-management-2025-06-27\"] : [];\n  let edits: Array<Record<string, unknown>> = [];\n\n  switch (contextProfile) {\n    case \"main\": {\n      const ctxMgmt = getContextManagement(model, contextOverrides);\n      betas = [...ctxMgmt.betas];\n      edits = ctxMgmt.config.edits;\n      break;\n    }\n    case \"subagent\":\n      if (isAnthropicModel) {\n        edits = [\n          {\n            type: \"clear_thinking_20251015\",\n            keep: { type: \"thinking_turns\", value: 1 },\n          },\n          {\n            type: \"clear_tool_uses_20250919\",\n            trigger: { type: \"input_tokens\", value: 60_000 },\n            keep: { type: \"tool_uses\", value: 2 },\n          },\n        ];\n        // Compaction for sub-agents on models that support it (lower threshold than main agent).\n        // Trigger at 80K — above the 60K clear_tool_uses threshold so clearing fires first.\n        const supportsCompaction = model.includes(\"opus-4-6\") || model.includes(\"sonnet-4-6\");\n        if (supportsCompaction) {\n          edits.push({\n            type: \"compact_20260112\",\n            trigger: { type: \"input_tokens\", value: 80_000 },\n            pause_after_compaction: true,\n            instructions: \"Summarize preserving: task goal, files found/modified, key findings, next steps.\",\n          });\n          betas.push(\"compact-2026-01-12\");\n        }\n      }\n      break;\n    case \"teammate\": {\n      const ctxMgmt = getContextManagement(model);\n      betas = [...ctxMgmt.betas];\n      edits = ctxMgmt.config.edits;\n      break;\n    }\n  }\n\n  // Thinking config\n  const thinkingCfg = getThinkingConfig(model, thinkingEnabled);\n  if (thinkingCfg.beta) betas.push(thinkingCfg.beta);\n\n  // Max tokens — caller is responsible for setting appropriate limits per context\n  const maxTokens = getMaxOutputTokens(model, maxOutputTokens);\n\n  // Build thinking param (ensure budget < maxTokens)\n  // Only include for Anthropic/Bedrock — Gemini/OpenAI handle thinking internally\n  let thinking: { type: string; budget_tokens?: number } | undefined;\n  if (isAnthropicModel && thinkingCfg.thinking.type !== \"disabled\") {\n    thinking = thinkingCfg.thinking.budget_tokens\n      ? { ...thinkingCfg.thinking, budget_tokens: Math.min(thinkingCfg.thinking.budget_tokens, maxTokens - 1) }\n      : thinkingCfg.thinking;\n  }\n\n  // Strip clear_thinking edits when thinking is disabled — API requires thinking\n  // to be enabled or adaptive for clear_thinking_20251015 to be accepted\n  if (!thinking) {\n    edits = edits.filter(e => e.type !== \"clear_thinking_20251015\");\n  }\n\n  return {\n    betas,\n    contextManagement: { edits },\n    thinking,\n    maxTokens,\n  };\n}\n\n// ============================================================================\n// SERVER PROXY CALLER\n// ============================================================================\n\nexport interface CallServerProxyConfig {\n  proxyUrl: string;\n  token: string;\n  model: string;\n  system: object[];\n  messages: object[];\n  tools: object[];\n  apiConfig: APIRequestConfig;\n  signal?: AbortSignal;\n  timeoutMs?: number;\n  fallbackModel?: string;\n  storeId?: string;  // Required for non-Anthropic providers (credential resolution)\n  onFallback?: (fromModel: string, toModel: string) => void;\n  onRetry?: (attempt: number, maxRetries: number, error: string) => void;\n  /** Called on 401 to refresh the auth token. Returns new token or null. */\n  onTokenRefresh?: () => Promise<string | null>;\n}\n\n/**\n * Call server proxy endpoint and return raw SSE stream.\n * Retries with exponential backoff on 429/500/529.\n */\nexport async function callServerProxy(\n  config: CallServerProxyConfig,\n): Promise<ReadableStream<Uint8Array>> {\n  const { proxyUrl, signal, timeoutMs } = config;\n  let tokenRefreshed = false;\n\n  for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {\n    try {\n      const body: Record<string, unknown> = {\n        mode: \"proxy\",\n        messages: config.messages,\n        system: config.system,\n        tools: config.tools,\n        model: config.model,\n        max_tokens: config.apiConfig.maxTokens,\n        stream: true,\n        betas: config.apiConfig.betas,\n        ...(config.apiConfig.contextManagement?.edits?.length\n          ? { context_management: config.apiConfig.contextManagement }\n          : {}),\n      };\n      if (config.apiConfig.thinking) {\n        body.thinking = config.apiConfig.thinking;\n        // Anthropic requires temperature=1.0 when thinking is enabled — enforce client-side\n        if (config.apiConfig.thinking.type !== \"disabled\") {\n          body.temperature = 1.0;\n        }\n      }\n      if (config.storeId) {\n        body.store_id = config.storeId;\n      }\n\n      // Prune old base64 images from conversation history before serializing.\n      // Keep images only in the last 2 messages to avoid multi-MB request bodies\n      // that can cause \"Invalid JSON\" errors from body truncation on the proxy.\n      if (Array.isArray(body.messages) && (body.messages as any[]).length > 4) {\n        const msgs = body.messages as any[];\n        const keepFrom = Math.max(0, msgs.length - 2);\n        for (let i = 0; i < keepFrom; i++) {\n          const msg = msgs[i];\n          if (!Array.isArray(msg?.content)) continue;\n          for (const block of msg.content) {\n            if (block.type === \"image\" && block.source?.type === \"base64\") {\n              block.type = \"text\";\n              block.text = \"[image]\";\n              delete block.source;\n            }\n            if (block.type === \"tool_result\" && Array.isArray(block.content)) {\n              for (const sub of block.content) {\n                if (sub.type === \"image\" && sub.source?.type === \"base64\") {\n                  sub.type = \"text\";\n                  sub.text = \"[image]\";\n                  delete sub.source;\n                }\n              }\n            }\n          }\n        }\n      }\n\n      const serialized = JSON.stringify(body);\n      const fetchOpts: RequestInit = {\n        method: \"POST\",\n        headers: {\n          \"Content-Type\": \"application/json\",\n          \"Authorization\": `Bearer ${config.token}`,\n          \"Content-Length\": String(Buffer.byteLength(serialized)),\n        },\n        body: serialized,\n        signal,\n      };\n\n      // Apply timeout only when the caller did NOT already supply a signal\n      let controller: AbortController | undefined;\n      let timeout: ReturnType<typeof setTimeout> | undefined;\n      if (timeoutMs && !signal) {\n        controller = new AbortController();\n        timeout = setTimeout(() => controller!.abort(), timeoutMs);\n        fetchOpts.signal = controller.signal;\n      }\n\n      try {\n        const response = await fetch(proxyUrl, fetchOpts);\n        if (timeout) clearTimeout(timeout);\n\n        if (response.ok && response.body) {\n          return response.body;\n        }\n\n        const errorBody = await response.text();\n\n        // Auto-refresh on 401 (token invalidated server-side before local expiry)\n        if (response.status === 401 && config.onTokenRefresh && !tokenRefreshed) {\n          const newToken = await config.onTokenRefresh();\n          if (newToken) {\n            tokenRefreshed = true;\n            config.token = newToken;\n            continue; // retry with new token\n          }\n        }\n\n        throw Object.assign(\n          new Error(`Proxy error (${response.status}): ${errorBody}`),\n          { status: response.status },\n        );\n      } catch (err) {\n        if (timeout) clearTimeout(timeout);\n        throw err;\n      }\n    } catch (err: unknown) {\n      if (signal?.aborted) throw err;\n\n      if (attempt < MAX_RETRIES && isRetryableError(err)) {\n        const errMsg = err instanceof Error ? err.message : String(err);\n        config.onRetry?.(attempt + 1, MAX_RETRIES, errMsg);\n        const delay = RETRY_BASE_DELAY_MS * Math.pow(2, attempt);\n\n        // Abort-aware backoff — ESC can interrupt the wait immediately\n        await new Promise<void>((resolve, reject) => {\n          if (signal?.aborted) { reject(signal.reason ?? new Error(\"Aborted\")); return; }\n          const timer = setTimeout(resolve, delay);\n          const onAbort = () => { clearTimeout(timer); reject(signal!.reason ?? new Error(\"Aborted\")); };\n          signal?.addEventListener(\"abort\", onAbort, { once: true });\n        });\n\n        // Fallback model on last retry\n        if (attempt === MAX_RETRIES - 1 && config.fallbackModel) {\n          const fromModel = config.model;\n          config.model = config.fallbackModel;\n          config.onFallback?.(fromModel, config.model);\n        }\n        continue;\n      }\n      throw err;\n    }\n  }\n\n  throw new Error(\"Failed to get response after retries\");\n}\n\n// ============================================================================\n// HELPERS — system prompt + caching setup\n// ============================================================================\n\n/**\n * Build system blocks with prompt caching.\n * Cached system prompt + optional dynamic cost context (after cache breakpoint).\n */\nexport function buildSystemBlocks(\n  systemPrompt: string,\n  costContext?: string,\n  enableCaching = true,\n): Array<Record<string, unknown>> {\n  const system: Array<Record<string, unknown>> = [];\n\n  if (enableCaching) {\n    system.push({\n      type: \"text\",\n      text: systemPrompt,\n      cache_control: { type: \"ephemeral\" },\n    });\n  } else {\n    system.push({ type: \"text\", text: systemPrompt });\n  }\n\n  if (costContext) {\n    system.push({ type: \"text\", text: costContext });\n  }\n\n  return system;\n}\n\n/**\n * Prepare messages and tools with prompt caching.\n * Skips cache_control injection for Gemini (uses implicit caching).\n */\nexport function prepareWithCaching<\n  T extends object = object,\n  M extends { role: string; content: unknown } = { role: string; content: unknown },\n>(\n  tools: T[],\n  messages: M[],\n  model?: string,\n): { tools: T[]; messages: M[] } {\n  if (model) {\n    const provider = getProvider(model);\n    if (provider === \"gemini\" || provider === \"openai\") {\n      return { tools, messages };\n    }\n  }\n  return addPromptCaching(tools, messages);\n}\n\n// ============================================================================\n// GEMINI CONTEXT TRIMMING\n// ============================================================================\n\n/**\n * Client-side context trimming for Gemini (no server-side context management).\n * When estimated input tokens exceed threshold, replaces old tool result contents\n * with \"[trimmed]\" to stay within Gemini's 1M context window.\n */\nexport function trimGeminiContext<T extends CompactableMessage>(\n  messages: T[],\n  estimatedTokens: number,\n  threshold = 950_000,\n  keepRecent = 5,\n): T[] {\n  if (estimatedTokens < threshold) return messages;\n\n  type Block = { type?: string; content?: unknown };\n  let toolResultCount = 0;\n  for (const msg of messages) {\n    if (Array.isArray(msg.content)) {\n      for (const block of msg.content as Block[]) {\n        if (block.type === \"tool_result\") toolResultCount++;\n      }\n    }\n  }\n\n  if (toolResultCount <= keepRecent) return messages;\n\n  const trimCount = toolResultCount - keepRecent;\n  let trimmed = 0;\n\n  return messages.map((msg) => {\n    if (!Array.isArray(msg.content)) return msg;\n    const content = (msg.content as Block[]).map((block) => {\n      if (block.type === \"tool_result\" && trimmed < trimCount) {\n        trimmed++;\n        if (typeof block.content === \"string\") {\n          return { ...block, content: \"[trimmed]\" };\n        } else if (Array.isArray(block.content)) {\n          return { ...block, content: [{ type: \"text\", text: \"[trimmed]\" }] };\n        }\n      }\n      return block;\n    });\n    return { ...msg, content } as T;\n  });\n}\n\n// ============================================================================\n// OPENAI CONTEXT TRIMMING\n// ============================================================================\n\n/**\n * Client-side context trimming for OpenAI (no server-side context management).\n * When estimated input tokens exceed threshold, replaces old tool result contents\n * with \"[trimmed]\" to stay within OpenAI's 200K context window.\n */\n// ============================================================================\n// AUDIO TRANSCRIPTION CLIENT\n// ============================================================================\n\n/**\n * Call server transcription endpoint (OpenAI Whisper via our Fly.io proxy).\n * Used by tool-dispatch when __AUDIO__ marker is detected in tool results.\n */\nexport async function callTranscribe(opts: {\n  proxyUrl: string;\n  token: string;\n  storeId: string;\n  audioBase64: string;\n  mediaType: string;\n}): Promise<string> {\n  const { proxyUrl, token, storeId, audioBase64, mediaType } = opts;\n\n  // Use same server URL as proxy, just different mode\n  const response = await fetch(proxyUrl, {\n    method: \"POST\",\n    headers: {\n      \"Content-Type\": \"application/json\",\n      Authorization: `Bearer ${token}`,\n    },\n    body: JSON.stringify({\n      mode: \"transcribe\",\n      audio_base64: audioBase64,\n      media_type: mediaType,\n      store_id: storeId,\n    }),\n  });\n\n  if (!response.ok) {\n    const errText = await response.text();\n    throw new Error(`Transcription failed (${response.status}): ${errText}`);\n  }\n\n  const data = (await response.json()) as { success: boolean; transcript?: string; error?: string };\n  if (!data.success || !data.transcript) {\n    throw new Error(data.error || \"Transcription returned no transcript\");\n  }\n\n  return data.transcript;\n}\n\n// ============================================================================\n// OPENAI CONTEXT TRIMMING\n// ============================================================================\n\nexport function trimOpenAIContext<T extends CompactableMessage>(\n  messages: T[],\n  estimatedTokens: number,\n  threshold = 190_000,\n  keepRecent = 5,\n): T[] {\n  if (estimatedTokens < threshold) return messages;\n\n  type Block = { type?: string; content?: unknown };\n  let toolResultCount = 0;\n  for (const msg of messages) {\n    if (Array.isArray(msg.content)) {\n      for (const block of msg.content as Block[]) {\n        if (block.type === \"tool_result\") toolResultCount++;\n      }\n    }\n  }\n\n  if (toolResultCount <= keepRecent) return messages;\n\n  const trimCount = toolResultCount - keepRecent;\n  let trimmed = 0;\n\n  return messages.map((msg) => {\n    if (!Array.isArray(msg.content)) return msg;\n    const content = (msg.content as Block[]).map((block) => {\n      if (block.type === \"tool_result\" && trimmed < trimCount) {\n        trimmed++;\n        if (typeof block.content === \"string\") {\n          return { ...block, content: \"[trimmed]\" };\n        } else if (Array.isArray(block.content)) {\n          return { ...block, content: [{ type: \"text\", text: \"[trimmed]\" }] };\n        }\n      }\n      return block;\n    });\n    return { ...msg, content } as T;\n  });\n}\n\n// ============================================================================\n// PROVIDER COMPACTION CLIENT\n// ============================================================================\n\nconst COMPACTION_TIMEOUT_MS = 30_000;\n\n/**\n * Request conversation compaction from the server (Haiku summarization).\n * Used by non-Anthropic providers (OpenAI, Gemini) that lack native compaction.\n * Returns summary string or null on failure (caller keeps unsummarized history).\n */\nexport async function requestProviderCompaction(opts: {\n  proxyUrl: string;\n  token: string;\n  messages: object[];\n  systemPrompt: string;\n}): Promise<string | null> {\n  const { proxyUrl, token, messages, systemPrompt } = opts;\n\n  try {\n    const controller = new AbortController();\n    const timeout = setTimeout(() => controller.abort(), COMPACTION_TIMEOUT_MS);\n\n    try {\n      const response = await fetch(proxyUrl, {\n        method: \"POST\",\n        headers: {\n          \"Content-Type\": \"application/json\",\n          Authorization: `Bearer ${token}`,\n        },\n        body: JSON.stringify({\n          mode: \"compact\",\n          messages,\n          system_prompt: systemPrompt,\n        }),\n        signal: controller.signal,\n      });\n\n      clearTimeout(timeout);\n\n      if (!response.ok) return null;\n\n      const data = (await response.json()) as {\n        success: boolean;\n        compaction_content?: string;\n        error?: string;\n      };\n\n      return data.success && data.compaction_content ? data.compaction_content : null;\n    } finally {\n      clearTimeout(timeout);\n    }\n  } catch {\n    return null; // Graceful failure — caller keeps unsummarized history\n  }\n}\n"],"mappings":"AAAA;AACA;AACA;AACA;AACA;AACA;;AAEA,SACEA,oBAAoB,EACpBC,kBAAkB,EAClBC,iBAAiB,EACjBC,gBAAgB,EAChBC,gBAAgB,QACX,iBAAiB;AAExB,SAASC,WAAW,QAAQ,gBAAgB;AAG5C;AACA;AACA;;AAEA,MAAMC,WAAW,GAAG,CAAC;AACrB,MAAMC,mBAAmB,GAAG,IAAI;;AAEhC;AACA;AACA;;AAEA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASC,eAAeA,CAACC,IAM/B,EAAoB;EACnB,MAAM;IAAEC,KAAK;IAAEC,cAAc;IAAEC,eAAe,GAAG,KAAK;IAAEC,eAAe;IAAEC;EAAiB,CAAC,GAAGL,IAAI;;EAElG;EACA;EACA,MAAMM,QAAQ,GAAGV,WAAW,CAACK,KAAK,CAAC;EACnC,MAAMM,gBAAgB,GAAGD,QAAQ,KAAK,WAAW,IAAIA,QAAQ,KAAK,SAAS;EAC3E,IAAIE,KAAe,GAAGD,gBAAgB,GAAG,CAAC,+BAA+B,CAAC,GAAG,EAAE;EAC/E,IAAIE,KAAqC,GAAG,EAAE;EAE9C,QAAQP,cAAc;IACpB,KAAK,MAAM;MAAE;QACX,MAAMQ,OAAO,GAAGnB,oBAAoB,CAACU,KAAK,EAAEI,gBAAgB,CAAC;QAC7DG,KAAK,GAAG,CAAC,GAAGE,OAAO,CAACF,KAAK,CAAC;QAC1BC,KAAK,GAAGC,OAAO,CAACC,MAAM,CAACF,KAAK;QAC5B;MACF;IACA,KAAK,UAAU;MACb,IAAIF,gBAAgB,EAAE;QACpBE,KAAK,GAAG,CACN;UACEG,IAAI,EAAE,yBAAyB;UAC/BC,IAAI,EAAE;YAAED,IAAI,EAAE,gBAAgB;YAAEE,KAAK,EAAE;UAAE;QAC3C,CAAC,EACD;UACEF,IAAI,EAAE,0BAA0B;UAChCG,OAAO,EAAE;YAAEH,IAAI,EAAE,cAAc;YAAEE,KAAK,EAAE;UAAO,CAAC;UAChDD,IAAI,EAAE;YAAED,IAAI,EAAE,WAAW;YAAEE,KAAK,EAAE;UAAE;QACtC,CAAC,CACF;QACD;QACA;QACA,MAAME,kBAAkB,GAAGf,KAAK,CAACgB,QAAQ,CAAC,UAAU,CAAC,IAAIhB,KAAK,CAACgB,QAAQ,CAAC,YAAY,CAAC;QACrF,IAAID,kBAAkB,EAAE;UACtBP,KAAK,CAACS,IAAI,CAAC;YACTN,IAAI,EAAE,kBAAkB;YACxBG,OAAO,EAAE;cAAEH,IAAI,EAAE,cAAc;cAAEE,KAAK,EAAE;YAAO,CAAC;YAChDK,sBAAsB,EAAE,IAAI;YAC5BC,YAAY,EAAE;UAChB,CAAC,CAAC;UACFZ,KAAK,CAACU,IAAI,CAAC,oBAAoB,CAAC;QAClC;MACF;MACA;IACF,KAAK,UAAU;MAAE;QACf,MAAMR,OAAO,GAAGnB,oBAAoB,CAACU,KAAK,CAAC;QAC3CO,KAAK,GAAG,CAAC,GAAGE,OAAO,CAACF,KAAK,CAAC;QAC1BC,KAAK,GAAGC,OAAO,CAACC,MAAM,CAACF,KAAK;QAC5B;MACF;EACF;;EAEA;EACA,MAAMY,WAAW,GAAG5B,iBAAiB,CAACQ,KAAK,EAAEE,eAAe,CAAC;EAC7D,IAAIkB,WAAW,CAACC,IAAI,EAAEd,KAAK,CAACU,IAAI,CAACG,WAAW,CAACC,IAAI,CAAC;;EAElD;EACA,MAAMC,SAAS,GAAG/B,kBAAkB,CAACS,KAAK,EAAEG,eAAe,CAAC;;EAE5D;EACA;EACA,IAAIoB,QAA8D;EAClE,IAAIjB,gBAAgB,IAAIc,WAAW,CAACG,QAAQ,CAACZ,IAAI,KAAK,UAAU,EAAE;IAChEY,QAAQ,GAAGH,WAAW,CAACG,QAAQ,CAACC,aAAa,GACzC;MAAE,GAAGJ,WAAW,CAACG,QAAQ;MAAEC,aAAa,EAAEC,IAAI,CAACC,GAAG,CAACN,WAAW,CAACG,QAAQ,CAACC,aAAa,EAAEF,SAAS,GAAG,CAAC;IAAE,CAAC,GACvGF,WAAW,CAACG,QAAQ;EAC1B;;EAEA;EACA;EACA,IAAI,CAACA,QAAQ,EAAE;IACbf,KAAK,GAAGA,KAAK,CAACmB,MAAM,CAACC,CAAC,IAAIA,CAAC,CAACjB,IAAI,KAAK,yBAAyB,CAAC;EACjE;EAEA,OAAO;IACLJ,KAAK;IACLsB,iBAAiB,EAAE;MAAErB;IAAM,CAAC;IAC5Be,QAAQ;IACRD;EACF,CAAC;AACH;;AAEA;AACA;AACA;;AAoBA;AACA;AACA;AACA;AACA,OAAO,eAAeQ,eAAeA,CACnCpB,MAA6B,EACQ;EACrC,MAAM;IAAEqB,QAAQ;IAAEC,MAAM;IAAEC;EAAU,CAAC,GAAGvB,MAAM;EAC9C,IAAIwB,cAAc,GAAG,KAAK;EAE1B,KAAK,IAAIC,OAAO,GAAG,CAAC,EAAEA,OAAO,IAAIvC,WAAW,EAAEuC,OAAO,EAAE,EAAE;IACvD,IAAI;MACF,MAAMC,IAA6B,GAAG;QACpCC,IAAI,EAAE,OAAO;QACbC,QAAQ,EAAE5B,MAAM,CAAC4B,QAAQ;QACzBC,MAAM,EAAE7B,MAAM,CAAC6B,MAAM;QACrBC,KAAK,EAAE9B,MAAM,CAAC8B,KAAK;QACnBxC,KAAK,EAAEU,MAAM,CAACV,KAAK;QACnByC,UAAU,EAAE/B,MAAM,CAACgC,SAAS,CAACpB,SAAS;QACtCqB,MAAM,EAAE,IAAI;QACZpC,KAAK,EAAEG,MAAM,CAACgC,SAAS,CAACnC,KAAK;QAC7B,IAAIG,MAAM,CAACgC,SAAS,CAACb,iBAAiB,EAAErB,KAAK,EAAEoC,MAAM,GACjD;UAAEC,kBAAkB,EAAEnC,MAAM,CAACgC,SAAS,CAACb;QAAkB,CAAC,GAC1D,CAAC,CAAC;MACR,CAAC;MACD,IAAInB,MAAM,CAACgC,SAAS,CAACnB,QAAQ,EAAE;QAC7Ba,IAAI,CAACb,QAAQ,GAAGb,MAAM,CAACgC,SAAS,CAACnB,QAAQ;QACzC;QACA,IAAIb,MAAM,CAACgC,SAAS,CAACnB,QAAQ,CAACZ,IAAI,KAAK,UAAU,EAAE;UACjDyB,IAAI,CAACU,WAAW,GAAG,GAAG;QACxB;MACF;MACA,IAAIpC,MAAM,CAACqC,OAAO,EAAE;QAClBX,IAAI,CAACY,QAAQ,GAAGtC,MAAM,CAACqC,OAAO;MAChC;;MAEA;MACA;MACA;MACA,IAAIE,KAAK,CAACC,OAAO,CAACd,IAAI,CAACE,QAAQ,CAAC,IAAKF,IAAI,CAACE,QAAQ,CAAWM,MAAM,GAAG,CAAC,EAAE;QACvE,MAAMO,IAAI,GAAGf,IAAI,CAACE,QAAiB;QACnC,MAAMc,QAAQ,GAAG3B,IAAI,CAAC4B,GAAG,CAAC,CAAC,EAAEF,IAAI,CAACP,MAAM,GAAG,CAAC,CAAC;QAC7C,KAAK,IAAIU,CAAC,GAAG,CAAC,EAAEA,CAAC,GAAGF,QAAQ,EAAEE,CAAC,EAAE,EAAE;UACjC,MAAMC,GAAG,GAAGJ,IAAI,CAACG,CAAC,CAAC;UACnB,IAAI,CAACL,KAAK,CAACC,OAAO,CAACK,GAAG,EAAEC,OAAO,CAAC,EAAE;UAClC,KAAK,MAAMC,KAAK,IAAIF,GAAG,CAACC,OAAO,EAAE;YAC/B,IAAIC,KAAK,CAAC9C,IAAI,KAAK,OAAO,IAAI8C,KAAK,CAACC,MAAM,EAAE/C,IAAI,KAAK,QAAQ,EAAE;cAC7D8C,KAAK,CAAC9C,IAAI,GAAG,MAAM;cACnB8C,KAAK,CAACE,IAAI,GAAG,SAAS;cACtB,OAAOF,KAAK,CAACC,MAAM;YACrB;YACA,IAAID,KAAK,CAAC9C,IAAI,KAAK,aAAa,IAAIsC,KAAK,CAACC,OAAO,CAACO,KAAK,CAACD,OAAO,CAAC,EAAE;cAChE,KAAK,MAAMI,GAAG,IAAIH,KAAK,CAACD,OAAO,EAAE;gBAC/B,IAAII,GAAG,CAACjD,IAAI,KAAK,OAAO,IAAIiD,GAAG,CAACF,MAAM,EAAE/C,IAAI,KAAK,QAAQ,EAAE;kBACzDiD,GAAG,CAACjD,IAAI,GAAG,MAAM;kBACjBiD,GAAG,CAACD,IAAI,GAAG,SAAS;kBACpB,OAAOC,GAAG,CAACF,MAAM;gBACnB;cACF;YACF;UACF;QACF;MACF;MAEA,MAAMG,UAAU,GAAGC,IAAI,CAACC,SAAS,CAAC3B,IAAI,CAAC;MACvC,MAAM4B,SAAsB,GAAG;QAC7BC,MAAM,EAAE,MAAM;QACdC,OAAO,EAAE;UACP,cAAc,EAAE,kBAAkB;UAClC,eAAe,EAAE,UAAUxD,MAAM,CAACyD,KAAK,EAAE;UACzC,gBAAgB,EAAEC,MAAM,CAACC,MAAM,CAACC,UAAU,CAACT,UAAU,CAAC;QACxD,CAAC;QACDzB,IAAI,EAAEyB,UAAU;QAChB7B;MACF,CAAC;;MAED;MACA,IAAIuC,UAAuC;MAC3C,IAAIC,OAAkD;MACtD,IAAIvC,SAAS,IAAI,CAACD,MAAM,EAAE;QACxBuC,UAAU,GAAG,IAAIE,eAAe,CAAC,CAAC;QAClCD,OAAO,GAAGE,UAAU,CAAC,MAAMH,UAAU,CAAEI,KAAK,CAAC,CAAC,EAAE1C,SAAS,CAAC;QAC1D+B,SAAS,CAAChC,MAAM,GAAGuC,UAAU,CAACvC,MAAM;MACtC;MAEA,IAAI;QACF,MAAM4C,QAAQ,GAAG,MAAMC,KAAK,CAAC9C,QAAQ,EAAEiC,SAAS,CAAC;QACjD,IAAIQ,OAAO,EAAEM,YAAY,CAACN,OAAO,CAAC;QAElC,IAAII,QAAQ,CAACG,EAAE,IAAIH,QAAQ,CAACxC,IAAI,EAAE;UAChC,OAAOwC,QAAQ,CAACxC,IAAI;QACtB;QAEA,MAAM4C,SAAS,GAAG,MAAMJ,QAAQ,CAACjB,IAAI,CAAC,CAAC;;QAEvC;QACA,IAAIiB,QAAQ,CAACK,MAAM,KAAK,GAAG,IAAIvE,MAAM,CAACwE,cAAc,IAAI,CAAChD,cAAc,EAAE;UACvE,MAAMiD,QAAQ,GAAG,MAAMzE,MAAM,CAACwE,cAAc,CAAC,CAAC;UAC9C,IAAIC,QAAQ,EAAE;YACZjD,cAAc,GAAG,IAAI;YACrBxB,MAAM,CAACyD,KAAK,GAAGgB,QAAQ;YACvB,SAAS,CAAC;UACZ;QACF;QAEA,MAAMC,MAAM,CAACC,MAAM,CACjB,IAAIC,KAAK,CAAC,gBAAgBV,QAAQ,CAACK,MAAM,MAAMD,SAAS,EAAE,CAAC,EAC3D;UAAEC,MAAM,EAAEL,QAAQ,CAACK;QAAO,CAC5B,CAAC;MACH,CAAC,CAAC,OAAOM,GAAG,EAAE;QACZ,IAAIf,OAAO,EAAEM,YAAY,CAACN,OAAO,CAAC;QAClC,MAAMe,GAAG;MACX;IACF,CAAC,CAAC,OAAOA,GAAY,EAAE;MACrB,IAAIvD,MAAM,EAAEwD,OAAO,EAAE,MAAMD,GAAG;MAE9B,IAAIpD,OAAO,GAAGvC,WAAW,IAAIH,gBAAgB,CAAC8F,GAAG,CAAC,EAAE;QAClD,MAAME,MAAM,GAAGF,GAAG,YAAYD,KAAK,GAAGC,GAAG,CAACG,OAAO,GAAGtB,MAAM,CAACmB,GAAG,CAAC;QAC/D7E,MAAM,CAACiF,OAAO,GAAGxD,OAAO,GAAG,CAAC,EAAEvC,WAAW,EAAE6F,MAAM,CAAC;QAClD,MAAMG,KAAK,GAAG/F,mBAAmB,GAAG4B,IAAI,CAACoE,GAAG,CAAC,CAAC,EAAE1D,OAAO,CAAC;;QAExD;QACA,MAAM,IAAI2D,OAAO,CAAO,CAACC,OAAO,EAAEC,MAAM,KAAK;UAC3C,IAAIhE,MAAM,EAAEwD,OAAO,EAAE;YAAEQ,MAAM,CAAChE,MAAM,CAACiE,MAAM,IAAI,IAAIX,KAAK,CAAC,SAAS,CAAC,CAAC;YAAE;UAAQ;UAC9E,MAAMY,KAAK,GAAGxB,UAAU,CAACqB,OAAO,EAAEH,KAAK,CAAC;UACxC,MAAMO,OAAO,GAAGA,CAAA,KAAM;YAAErB,YAAY,CAACoB,KAAK,CAAC;YAAEF,MAAM,CAAChE,MAAM,CAAEiE,MAAM,IAAI,IAAIX,KAAK,CAAC,SAAS,CAAC,CAAC;UAAE,CAAC;UAC9FtD,MAAM,EAAEoE,gBAAgB,CAAC,OAAO,EAAED,OAAO,EAAE;YAAEE,IAAI,EAAE;UAAK,CAAC,CAAC;QAC5D,CAAC,CAAC;;QAEF;QACA,IAAIlE,OAAO,KAAKvC,WAAW,GAAG,CAAC,IAAIc,MAAM,CAAC4F,aAAa,EAAE;UACvD,MAAMC,SAAS,GAAG7F,MAAM,CAACV,KAAK;UAC9BU,MAAM,CAACV,KAAK,GAAGU,MAAM,CAAC4F,aAAa;UACnC5F,MAAM,CAAC8F,UAAU,GAAGD,SAAS,EAAE7F,MAAM,CAACV,KAAK,CAAC;QAC9C;QACA;MACF;MACA,MAAMuF,GAAG;IACX;EACF;EAEA,MAAM,IAAID,KAAK,CAAC,sCAAsC,CAAC;AACzD;;AAEA;AACA;AACA;;AAEA;AACA;AACA;AACA;AACA,OAAO,SAASmB,iBAAiBA,CAC/BC,YAAoB,EACpBC,WAAoB,EACpBC,aAAa,GAAG,IAAI,EACY;EAChC,MAAMrE,MAAsC,GAAG,EAAE;EAEjD,IAAIqE,aAAa,EAAE;IACjBrE,MAAM,CAACtB,IAAI,CAAC;MACVN,IAAI,EAAE,MAAM;MACZgD,IAAI,EAAE+C,YAAY;MAClBG,aAAa,EAAE;QAAElG,IAAI,EAAE;MAAY;IACrC,CAAC,CAAC;EACJ,CAAC,MAAM;IACL4B,MAAM,CAACtB,IAAI,CAAC;MAAEN,IAAI,EAAE,MAAM;MAAEgD,IAAI,EAAE+C;IAAa,CAAC,CAAC;EACnD;EAEA,IAAIC,WAAW,EAAE;IACfpE,MAAM,CAACtB,IAAI,CAAC;MAAEN,IAAI,EAAE,MAAM;MAAEgD,IAAI,EAAEgD;IAAY,CAAC,CAAC;EAClD;EAEA,OAAOpE,MAAM;AACf;;AAEA;AACA;AACA;AACA;AACA,OAAO,SAASuE,kBAAkBA,CAIhCtE,KAAU,EACVF,QAAa,EACbtC,KAAc,EACiB;EAC/B,IAAIA,KAAK,EAAE;IACT,MAAMK,QAAQ,GAAGV,WAAW,CAACK,KAAK,CAAC;IACnC,IAAIK,QAAQ,KAAK,QAAQ,IAAIA,QAAQ,KAAK,QAAQ,EAAE;MAClD,OAAO;QAAEmC,KAAK;QAAEF;MAAS,CAAC;IAC5B;EACF;EACA,OAAO5C,gBAAgB,CAAC8C,KAAK,EAAEF,QAAQ,CAAC;AAC1C;;AAEA;AACA;AACA;;AAEA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASyE,iBAAiBA,CAC/BzE,QAAa,EACb0E,eAAuB,EACvBC,SAAS,GAAG,OAAO,EACnBC,UAAU,GAAG,CAAC,EACT;EACL,IAAIF,eAAe,GAAGC,SAAS,EAAE,OAAO3E,QAAQ;EAGhD,IAAI6E,eAAe,GAAG,CAAC;EACvB,KAAK,MAAM5D,GAAG,IAAIjB,QAAQ,EAAE;IAC1B,IAAIW,KAAK,CAACC,OAAO,CAACK,GAAG,CAACC,OAAO,CAAC,EAAE;MAC9B,KAAK,MAAMC,KAAK,IAAIF,GAAG,CAACC,OAAO,EAAa;QAC1C,IAAIC,KAAK,CAAC9C,IAAI,KAAK,aAAa,EAAEwG,eAAe,EAAE;MACrD;IACF;EACF;EAEA,IAAIA,eAAe,IAAID,UAAU,EAAE,OAAO5E,QAAQ;EAElD,MAAM8E,SAAS,GAAGD,eAAe,GAAGD,UAAU;EAC9C,IAAIG,OAAO,GAAG,CAAC;EAEf,OAAO/E,QAAQ,CAACgF,GAAG,CAAE/D,GAAG,IAAK;IAC3B,IAAI,CAACN,KAAK,CAACC,OAAO,CAACK,GAAG,CAACC,OAAO,CAAC,EAAE,OAAOD,GAAG;IAC3C,MAAMC,OAAO,GAAID,GAAG,CAACC,OAAO,CAAa8D,GAAG,CAAE7D,KAAK,IAAK;MACtD,IAAIA,KAAK,CAAC9C,IAAI,KAAK,aAAa,IAAI0G,OAAO,GAAGD,SAAS,EAAE;QACvDC,OAAO,EAAE;QACT,IAAI,OAAO5D,KAAK,CAACD,OAAO,KAAK,QAAQ,EAAE;UACrC,OAAO;YAAE,GAAGC,KAAK;YAAED,OAAO,EAAE;UAAY,CAAC;QAC3C,CAAC,MAAM,IAAIP,KAAK,CAACC,OAAO,CAACO,KAAK,CAACD,OAAO,CAAC,EAAE;UACvC,OAAO;YAAE,GAAGC,KAAK;YAAED,OAAO,EAAE,CAAC;cAAE7C,IAAI,EAAE,MAAM;cAAEgD,IAAI,EAAE;YAAY,CAAC;UAAE,CAAC;QACrE;MACF;MACA,OAAOF,KAAK;IACd,CAAC,CAAC;IACF,OAAO;MAAE,GAAGF,GAAG;MAAEC;IAAQ,CAAC;EAC5B,CAAC,CAAC;AACJ;;AAEA;AACA;AACA;;AAEA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;;AAEA;AACA;AACA;AACA;AACA,OAAO,eAAe+D,cAAcA,CAACxH,IAMpC,EAAmB;EAClB,MAAM;IAAEgC,QAAQ;IAAEoC,KAAK;IAAEpB,OAAO;IAAEyE,WAAW;IAAEC;EAAU,CAAC,GAAG1H,IAAI;;EAEjE;EACA,MAAM6E,QAAQ,GAAG,MAAMC,KAAK,CAAC9C,QAAQ,EAAE;IACrCkC,MAAM,EAAE,MAAM;IACdC,OAAO,EAAE;MACP,cAAc,EAAE,kBAAkB;MAClCwD,aAAa,EAAE,UAAUvD,KAAK;IAChC,CAAC;IACD/B,IAAI,EAAE0B,IAAI,CAACC,SAAS,CAAC;MACnB1B,IAAI,EAAE,YAAY;MAClBsF,YAAY,EAAEH,WAAW;MACzBI,UAAU,EAAEH,SAAS;MACrBzE,QAAQ,EAAED;IACZ,CAAC;EACH,CAAC,CAAC;EAEF,IAAI,CAAC6B,QAAQ,CAACG,EAAE,EAAE;IAChB,MAAM8C,OAAO,GAAG,MAAMjD,QAAQ,CAACjB,IAAI,CAAC,CAAC;IACrC,MAAM,IAAI2B,KAAK,CAAC,yBAAyBV,QAAQ,CAACK,MAAM,MAAM4C,OAAO,EAAE,CAAC;EAC1E;EAEA,MAAMC,IAAI,GAAI,MAAMlD,QAAQ,CAACmD,IAAI,CAAC,CAA+D;EACjG,IAAI,CAACD,IAAI,CAACE,OAAO,IAAI,CAACF,IAAI,CAACG,UAAU,EAAE;IACrC,MAAM,IAAI3C,KAAK,CAACwC,IAAI,CAACI,KAAK,IAAI,sCAAsC,CAAC;EACvE;EAEA,OAAOJ,IAAI,CAACG,UAAU;AACxB;;AAEA;AACA;AACA;;AAEA,OAAO,SAASE,iBAAiBA,CAC/B7F,QAAa,EACb0E,eAAuB,EACvBC,SAAS,GAAG,OAAO,EACnBC,UAAU,GAAG,CAAC,EACT;EACL,IAAIF,eAAe,GAAGC,SAAS,EAAE,OAAO3E,QAAQ;EAGhD,IAAI6E,eAAe,GAAG,CAAC;EACvB,KAAK,MAAM5D,GAAG,IAAIjB,QAAQ,EAAE;IAC1B,IAAIW,KAAK,CAACC,OAAO,CAACK,GAAG,CAACC,OAAO,CAAC,EAAE;MAC9B,KAAK,MAAMC,KAAK,IAAIF,GAAG,CAACC,OAAO,EAAa;QAC1C,IAAIC,KAAK,CAAC9C,IAAI,KAAK,aAAa,EAAEwG,eAAe,EAAE;MACrD;IACF;EACF;EAEA,IAAIA,eAAe,IAAID,UAAU,EAAE,OAAO5E,QAAQ;EAElD,MAAM8E,SAAS,GAAGD,eAAe,GAAGD,UAAU;EAC9C,IAAIG,OAAO,GAAG,CAAC;EAEf,OAAO/E,QAAQ,CAACgF,GAAG,CAAE/D,GAAG,IAAK;IAC3B,IAAI,CAACN,KAAK,CAACC,OAAO,CAACK,GAAG,CAACC,OAAO,CAAC,EAAE,OAAOD,GAAG;IAC3C,MAAMC,OAAO,GAAID,GAAG,CAACC,OAAO,CAAa8D,GAAG,CAAE7D,KAAK,IAAK;MACtD,IAAIA,KAAK,CAAC9C,IAAI,KAAK,aAAa,IAAI0G,OAAO,GAAGD,SAAS,EAAE;QACvDC,OAAO,EAAE;QACT,IAAI,OAAO5D,KAAK,CAACD,OAAO,KAAK,QAAQ,EAAE;UACrC,OAAO;YAAE,GAAGC,KAAK;YAAED,OAAO,EAAE;UAAY,CAAC;QAC3C,CAAC,MAAM,IAAIP,KAAK,CAACC,OAAO,CAACO,KAAK,CAACD,OAAO,CAAC,EAAE;UACvC,OAAO;YAAE,GAAGC,KAAK;YAAED,OAAO,EAAE,CAAC;cAAE7C,IAAI,EAAE,MAAM;cAAEgD,IAAI,EAAE;YAAY,CAAC;UAAE,CAAC;QACrE;MACF;MACA,OAAOF,KAAK;IACd,CAAC,CAAC;IACF,OAAO;MAAE,GAAGF,GAAG;MAAEC;IAAQ,CAAC;EAC5B,CAAC,CAAC;AACJ;;AAEA;AACA;AACA;;AAEA,MAAM4E,qBAAqB,GAAG,MAAM;;AAEpC;AACA;AACA;AACA;AACA;AACA,OAAO,eAAeC,yBAAyBA,CAACtI,IAK/C,EAA0B;EACzB,MAAM;IAAEgC,QAAQ;IAAEoC,KAAK;IAAE7B,QAAQ;IAAEoE;EAAa,CAAC,GAAG3G,IAAI;EAExD,IAAI;IACF,MAAMwE,UAAU,GAAG,IAAIE,eAAe,CAAC,CAAC;IACxC,MAAMD,OAAO,GAAGE,UAAU,CAAC,MAAMH,UAAU,CAACI,KAAK,CAAC,CAAC,EAAEyD,qBAAqB,CAAC;IAE3E,IAAI;MACF,MAAMxD,QAAQ,GAAG,MAAMC,KAAK,CAAC9C,QAAQ,EAAE;QACrCkC,MAAM,EAAE,MAAM;QACdC,OAAO,EAAE;UACP,cAAc,EAAE,kBAAkB;UAClCwD,aAAa,EAAE,UAAUvD,KAAK;QAChC,CAAC;QACD/B,IAAI,EAAE0B,IAAI,CAACC,SAAS,CAAC;UACnB1B,IAAI,EAAE,SAAS;UACfC,QAAQ;UACRgG,aAAa,EAAE5B;QACjB,CAAC,CAAC;QACF1E,MAAM,EAAEuC,UAAU,CAACvC;MACrB,CAAC,CAAC;MAEF8C,YAAY,CAACN,OAAO,CAAC;MAErB,IAAI,CAACI,QAAQ,CAACG,EAAE,EAAE,OAAO,IAAI;MAE7B,MAAM+C,IAAI,GAAI,MAAMlD,QAAQ,CAACmD,IAAI,CAAC,CAIjC;MAED,OAAOD,IAAI,CAACE,OAAO,IAAIF,IAAI,CAACS,kBAAkB,GAAGT,IAAI,CAACS,kBAAkB,GAAG,IAAI;IACjF,CAAC,SAAS;MACRzD,YAAY,CAACN,OAAO,CAAC;IACvB;EACF,CAAC,CAAC,MAAM;IACN,OAAO,IAAI,CAAC,CAAC;EACf;AACF","ignoreList":[]}

package/dist/shared/api-client.test.js

@@ -0,0 +1,228 @@
+import { describe, it, expect } from "vitest";
+import { buildAPIRequest } from "./api-client.js";
+
+// ============================================================================
+// buildAPIRequest
+// ============================================================================
+
+describe("buildAPIRequest", () => {
+  // ---------- main profile ----------
+
+  describe("main profile", () => {
+    it("returns compact + clear betas for opus-4-6", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main"
+      });
+      expect(config.betas).toContain("context-management-2025-06-27");
+      expect(config.betas).toContain("compact-2026-01-12");
+    });
+    it("returns compact + clear betas for sonnet-4-6", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "main"
+      });
+      expect(config.betas).toContain("context-management-2025-06-27");
+      expect(config.betas).toContain("compact-2026-01-12");
+    });
+    it("includes compact edit at 167K and clear edit at 100K for opus-4-6", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main"
+      });
+      const compactEdit = config.contextManagement.edits.find(e => e.type === "compact_20260112");
+      const clearEdit = config.contextManagement.edits.find(e => e.type === "clear_tool_uses_20250919");
+      expect(compactEdit).toBeDefined();
+      expect(compactEdit.trigger.value).toBe(167_000);
+      expect(clearEdit).toBeDefined();
+      expect(clearEdit.trigger.value).toBe(100_000);
+      expect(clearEdit.keep.value).toBe(3);
+    });
+    it("includes compact + clear_tool_uses edits for sonnet-4-6 (thinking disabled)", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "main"
+      });
+      // With thinking disabled, clear_thinking is filtered out, leaving compact + clear_tool_uses
+      expect(config.contextManagement.edits).toHaveLength(2);
+      expect(config.contextManagement.edits[0].type).toBe("compact_20260112");
+      expect(config.contextManagement.edits[1].type).toBe("clear_tool_uses_20250919");
+    });
+    it("defaults maxTokens to 32000", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main"
+      });
+      expect(config.maxTokens).toBe(32000);
+    });
+    it("respects maxOutputTokens override", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main",
+        maxOutputTokens: 4096
+      });
+      expect(config.maxTokens).toBe(4096);
+    });
+    it("caps maxOutputTokens at model maximum", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main",
+        maxOutputTokens: 200000
+      });
+      // 200000 exceeds opus model max of 128000, so capped
+      expect(config.maxTokens).toBe(128000);
+    });
+  });
+
+  // ---------- subagent profile ----------
+
+  describe("subagent profile", () => {
+    it("uses clear at 60K/keep 2 for Anthropic models", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "subagent"
+      });
+      // Subagent profile now includes clear_thinking + clear_tool_uses (+ compact for supported models)
+      expect(config.contextManagement.edits.length).toBeGreaterThanOrEqual(2);
+      const edit = config.contextManagement.edits.find(e => e.type === "clear_tool_uses_20250919");
+      expect(edit).toBeDefined();
+      expect(edit.trigger.value).toBe(60_000);
+      expect(edit.keep.value).toBe(2);
+    });
+    it("defaults maxTokens to 32000 (subagent cap)", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "subagent"
+      });
+      expect(config.maxTokens).toBe(32000);
+    });
+    it("allows explicit maxOutputTokens to override subagent default", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "subagent",
+        maxOutputTokens: 4096
+      });
+      expect(config.maxTokens).toBe(4096);
+    });
+    it("has context-management beta for Anthropic models", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "subagent"
+      });
+      expect(config.betas).toContain("context-management-2025-06-27");
+    });
+  });
+
+  // ---------- teammate profile ----------
+
+  describe("teammate profile", () => {
+    it("uses same context management as main for opus-4-6", () => {
+      const main = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main"
+      });
+      const teammate = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "teammate"
+      });
+      expect(teammate.betas).toEqual(main.betas);
+      expect(teammate.contextManagement.edits).toEqual(main.contextManagement.edits);
+    });
+    it("defaults maxTokens to 32000 (not subagent cap)", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "teammate"
+      });
+      expect(config.maxTokens).toBe(32000);
+    });
+  });
+
+  // ---------- thinking config ----------
+
+  describe("thinking integration", () => {
+    it("does not include thinking when disabled", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main",
+        thinkingEnabled: false
+      });
+      expect(config.thinking).toBeUndefined();
+    });
+    it("defaults thinkingEnabled to false", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main"
+      });
+      expect(config.thinking).toBeUndefined();
+    });
+    it("includes adaptive thinking beta for opus-4-6 when enabled", () => {
+      const config = buildAPIRequest({
+        model: "claude-opus-4-6",
+        contextProfile: "main",
+        thinkingEnabled: true
+      });
+      expect(config.betas).toContain("adaptive-thinking-2026-01-28");
+      expect(config.thinking).toBeDefined();
+      expect(config.thinking.type).toBe("adaptive");
+      expect(config.thinking.budget_tokens).toBeUndefined();
+    });
+    it("includes adaptive thinking beta for sonnet-4-6 when enabled", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-6",
+        contextProfile: "main",
+        thinkingEnabled: true
+      });
+      expect(config.betas).toContain("adaptive-thinking-2026-01-28");
+      expect(config.thinking).toBeDefined();
+      expect(config.thinking.type).toBe("adaptive");
+      expect(config.thinking.budget_tokens).toBeUndefined();
+    });
+    it("uses fixed budget thinking for older sonnet models", () => {
+      const config = buildAPIRequest({
+        model: "claude-sonnet-4-20250514",
+        contextProfile: "main",
+        thinkingEnabled: true
+      });
+      // Older sonnet uses fixed budget, not adaptive
+      expect(config.thinking).toBeDefined();
+      expect(config.thinking.type).toBe("enabled");
+      expect(config.thinking.budget_tokens).toBeDefined();
+      expect(config.thinking.budget_tokens).toBeLessThan(config.maxTokens);
+    });
+  });
+
+  // ---------- Gemini (non-Anthropic) models ----------
+
+  describe("non-Anthropic models (Gemini)", () => {
+    it("returns empty betas for gemini models on main profile", () => {
+      const config = buildAPIRequest({
+        model: "gemini-3-pro-preview",
+        contextProfile: "main"
+      });
+      expect(config.betas).toEqual([]);
+    });
+    it("returns empty edits for gemini models on main profile", () => {
+      const config = buildAPIRequest({
+        model: "gemini-3-pro-preview",
+        contextProfile: "main"
+      });
+      expect(config.contextManagement.edits).toEqual([]);
+    });
+    it("returns empty betas for gemini models on subagent profile", () => {
+      const config = buildAPIRequest({
+        model: "gemini-2.5-flash",
+        contextProfile: "subagent"
+      });
+      expect(config.betas).toEqual([]);
+      expect(config.contextManagement.edits).toEqual([]);
+    });
+    it("disables thinking for gemini models even when enabled", () => {
+      const config = buildAPIRequest({
+        model: "gemini-3-pro-preview",
+        contextProfile: "main",
+        thinkingEnabled: true
+      });
+      expect(config.thinking).toBeUndefined();
+    });
+  });
+});

package/dist/shared/compaction-thinking.test.js

@@ -0,0 +1,315 @@
+/**
+ * Tests for shared/compaction.ts — thinking stripping, combined scenarios, bytesRemoved
+ *
+ * Covers:
+ * - Thinking block stripping
+ * - Combined real-world scenarios
+ * - bytesRemoved accuracy tracking
+ */
+
+import { describe, it, expect } from "vitest";
+import { preCompact } from "./compaction.js";
+
+// ============================================================================
+// HELPERS
+// ============================================================================
+
+/** Build a user message with a tool_result block */
+function toolResultMsg(toolUseId, content) {
+  return {
+    role: "user",
+    content: [{
+      type: "tool_result",
+      tool_use_id: toolUseId,
+      content
+    }]
+  };
+}
+
+/** Build an assistant message with text */
+function assistantMsg(text) {
+  return {
+    role: "assistant",
+    content: [{
+      type: "text",
+      text
+    }]
+  };
+}
+
+/** Build an assistant message with thinking + text */
+function assistantWithThinking(text, thinking) {
+  return {
+    role: "assistant",
+    content: [{
+      type: "thinking",
+      thinking,
+      signature: "sig"
+    }, {
+      type: "text",
+      text
+    }]
+  };
+}
+
+/** Generate a string of N characters */
+function bigStr(n) {
+  return "x".repeat(n);
+}
+
+// ============================================================================
+// TESTS
+// ============================================================================
+
+describe("preCompact", () => {
+  // --------------------------------------------------------------------------
+  // Thinking block stripping
+  // --------------------------------------------------------------------------
+  describe("thinking block stripping", () => {
+    it("strips thinking blocks older than 3 turns from end", () => {
+      const msgs = [assistantWithThinking("old answer", "old thinking"),
+      // msg 0
+      {
+        role: "user",
+        content: "q1"
+      },
+      // msg 1
+      assistantWithThinking("middle answer", "mid thinking"),
+      // msg 2
+      {
+        role: "user",
+        content: "q2"
+      },
+      // msg 3
+      assistantWithThinking("recent1", "recent-think-1"),
+      // msg 4
+      {
+        role: "user",
+        content: "q3"
+      },
+      // msg 5
+      assistantWithThinking("recent2", "recent-think-2"),
+      // msg 6
+      {
+        role: "user",
+        content: "q4"
+      } // msg 7
+      ];
+      // 8 messages: cutoff = max(0, 8-6) = 2. Messages 0-1 are in "old" zone.
+      const {
+        messages,
+        bytesRemoved
+      } = preCompact(msgs);
+
+      // msg 0 (assistant) — thinking should be stripped
+      const msg0Content = messages[0].content;
+      expect(msg0Content.some(b => b.type === "thinking")).toBe(false);
+      expect(msg0Content.some(b => b.type === "text")).toBe(true);
+
+      // msg 2 (assistant at idx >= cutoff=2) — thinking preserved
+      const msg2Content = messages[2].content;
+      expect(msg2Content.some(b => b.type === "thinking")).toBe(true);
+      expect(bytesRemoved).toBeGreaterThan(0);
+    });
+    it("preserves thinking blocks when stripThinking is false", () => {
+      const msgs = [assistantWithThinking("old", "think-old"), {
+        role: "user",
+        content: "q"
+      }, assistantWithThinking("new", "think-new"), {
+        role: "user",
+        content: "q2"
+      }, assistantWithThinking("newest", "think-newest"), {
+        role: "user",
+        content: "q3"
+      }, assistantWithThinking("end", "think-end"), {
+        role: "user",
+        content: "q4"
+      }];
+      const {
+        messages,
+        bytesRemoved
+      } = preCompact(msgs, {
+        stripThinking: false
+      });
+      // All thinking blocks should be preserved
+      for (const msg of messages) {
+        if (msg.role === "assistant" && Array.isArray(msg.content)) {
+          const hasThinking = msg.content.some(b => b.type === "thinking");
+          expect(hasThinking).toBe(true);
+        }
+      }
+      // bytesRemoved might still be >0 from other passes, but thinking not counted
+    });
+    it("preserves assistant message even if all content is thinking (no empty messages)", () => {
+      const msgs = [{
+        role: "assistant",
+        content: [{
+          type: "thinking",
+          thinking: "only thinking",
+          signature: "sig"
+        }]
+      }, {
+        role: "user",
+        content: "q"
+      }, assistantMsg("recent1"), {
+        role: "user",
+        content: "q2"
+      }, assistantMsg("recent2"), {
+        role: "user",
+        content: "q3"
+      }, assistantMsg("recent3"), {
+        role: "user",
+        content: "q4"
+      }];
+      const {
+        messages
+      } = preCompact(msgs);
+      // msg 0 is in the old zone (cutoff = 2), but has only thinking
+      // Should keep the message as-is (the `filtered.length === 0` guard)
+      expect(messages[0].content).toBeTruthy();
+    });
+  });
+
+  // --------------------------------------------------------------------------
+  // Combined scenarios (real-world patterns)
+  // --------------------------------------------------------------------------
+  describe("combined scenarios", () => {
+    it("handles a realistic 5-turn subagent conversation", () => {
+      const fileContent = bigStr(10000); // Large file read
+      const msgs = [
+      // Turn 1: glob + read
+      {
+        role: "user",
+        content: [{
+          type: "tool_result",
+          tool_use_id: "glob-1",
+          content: JSON.stringify(["a.ts", "b.ts"])
+        }, {
+          type: "tool_result",
+          tool_use_id: "read-1",
+          content: fileContent
+        }]
+      }, assistantWithThinking("Found the file", "Let me analyze..."),
+      // Turn 2: grep
+      {
+        role: "user",
+        content: [{
+          type: "tool_result",
+          tool_use_id: "grep-1",
+          content: bigStr(3000)
+        }]
+      }, assistantWithThinking("Found matches", "Grep found..."),
+      // Turn 3: read same file again (duplicate)
+      {
+        role: "user",
+        content: [{
+          type: "tool_result",
+          tool_use_id: "read-2",
+          content: fileContent
+        }]
+      }, assistantMsg("Checking again..."),
+      // Turn 4: edit
+      {
+        role: "user",
+        content: [{
+          type: "tool_result",
+          tool_use_id: "edit-1",
+          content: "File edited successfully"
+        }]
+      }, assistantMsg("Done")];
+      const {
+        messages,
+        bytesRemoved
+      } = preCompact(msgs);
+
+      // Large file read in turn 1 should be trimmed (has followup)
+      const turn1 = messages[0].content;
+      const readBlock = turn1.find(b => b.tool_use_id === "read-1");
+      expect(readBlock.content).toContain("[Tool result:");
+
+      // Duplicate file read in turn 3 should show "Duplicate"
+      const turn3 = messages[4].content;
+      // It was already trimmed (>2048) AND is a duplicate — trimming happens first
+      expect(turn3[0].content).toContain("[Tool result:");
+
+      // Old thinking (turn 1, msg idx 1) should be stripped
+      // cutoff = max(0, 8 - 6) = 2. Msg 1 is at idx 1 < 2.
+      const turn1Assistant = messages[1].content;
+      expect(turn1Assistant.some(b => b.type === "thinking")).toBe(false);
+
+      // Recent thinking (turn 2, msg idx 3) should be preserved (idx 3 >= cutoff 2)
+      const turn2Assistant = messages[3].content;
+      expect(turn2Assistant.some(b => b.type === "thinking")).toBe(true);
+
+      // Small edit result should be untouched
+      const turn4 = messages[6].content;
+      expect(turn4[0].content).toBe("File edited successfully");
+
+      // Should have removed significant bytes
+      expect(bytesRemoved).toBeGreaterThan(10000);
+    });
+    it("stress test: 20 turns with mixed large/small results", () => {
+      const msgs = [];
+      for (let i = 0; i < 20; i++) {
+        msgs.push({
+          role: "user",
+          content: [{
+            type: "tool_result",
+            tool_use_id: `t${i}`,
+            content: i % 3 === 0 ? bigStr(5000) : bigStr(100)
+          }]
+        });
+        msgs.push(assistantMsg(`Turn ${i} done`));
+      }
+      const {
+        messages,
+        bytesRemoved
+      } = preCompact(msgs);
+      expect(messages).toHaveLength(40);
+      expect(bytesRemoved).toBeGreaterThan(0);
+
+      // All large results (except possibly the very last) should be trimmed
+      for (let i = 0; i < 19; i++) {
+        if (i % 3 === 0) {
+          const block = messages[i * 2].content[0];
+          expect(block.content).toContain("[Tool result:");
+        }
+      }
+    });
+  });
+
+  // --------------------------------------------------------------------------
+  // bytesRemoved accuracy
+  // --------------------------------------------------------------------------
+  describe("bytesRemoved tracking", () => {
+    it("accurately tracks bytes removed from large tool results", () => {
+      const size = 5000;
+      const msgs = [toolResultMsg("t1", bigStr(size)), assistantMsg("ok")];
+      const {
+        bytesRemoved
+      } = preCompact(msgs);
+      // Should be approximately size - 100 (the placeholder is ~100 chars shorter)
+      expect(bytesRemoved).toBe(size - 100);
+    });
+    it("accumulates bytes from multiple sources", () => {
+      const msgs = [toolResultMsg("t1", bigStr(3000)),
+      // trimmed: ~2900 removed
+      assistantWithThinking("old", bigStr(500)),
+      // thinking stripped: 500 removed
+      toolResultMsg("t2", bigStr(4000)),
+      // trimmed: ~3900 removed
+      assistantMsg("mid"), assistantMsg("recent1"), {
+        role: "user",
+        content: "q"
+      }, assistantMsg("recent2"), {
+        role: "user",
+        content: "q2"
+      }];
+      const {
+        bytesRemoved
+      } = preCompact(msgs);
+      // Should include both tool result trimming and thinking stripping
+      expect(bytesRemoved).toBeGreaterThan(5000);
+    });
+  });
+});