whale-code 6.5.5 → 6.5.7

package/dist/shared/agent-core-types.js

@@ -0,0 +1,265 @@
+/**
+ * Agent Core Types — shared type definitions
+ * Extracted from agent-core.ts for modularity.
+ */
+
+// ============================================================================
+// TYPES
+// ============================================================================
+
+/**
+ * Controls which tools the model can/must use:
+ * - "auto": Model decides whether to use tools (default)
+ * - "any": Model must use at least one tool
+ * - "none": Model must not use any tools (text-only response)
+ * - { type: "tool", name: string }: Model must use the specified tool
+ */
+
+/** Overrides from ai_agent_config.context_config JSONB — all optional, falls back to defaults */
+
+/** Context config shape from ai_agent_config.context_config JSONB */
+
+// ============================================================================
+// CONSTANTS
+// ============================================================================
+
+export const AGENT_DEFAULTS = {
+  maxTurns: 25,
+  temperature: 0.7,
+  maxTokens: 16384,
+  maxConcurrentTools: 4,
+  enableDelegation: true,
+  maxDurationMs: 120_000,
+  maxDurationMsWorkflow: 300_000,
+  enableModelRouting: false,
+  maxMessageChars: 100_000,
+  maxHistoryChars: 500_000,
+  maxToolResultChars: 100_000,
+  subagentMaxTokens: 16384,
+  subagentMaxTurns: 15,
+  subagentTemperature: 0.7,
+  maxContinuations: 5,
+  subagentMaxContinuations: 3,
+  subagentMaxRetries: 2,
+  teammateMaxTotalTurns: 100,
+  teamTimeoutMs: 600_000,
+  workerStallMs: 120_000,
+  teamReviewMaxTokens: 16384,
+  apiTimeoutMs: 120_000,
+  toolTimeoutMs: 120_000,
+  toolResultTruncateChars: 50_000,
+  compactionTriggerTokens: 167_000,
+  // Loop detector
+  loopIdenticalCallLimit: 4,
+  loopConsecutiveErrorLimit: 5,
+  loopTurnErrorLimit: 10,
+  loopWindow: 20,
+  loopSessionToolErrorLimit: 25,
+  loopConsecutiveFailedTurnLimit: 3,
+  loopFileReadLimit: 8,
+  loopBatchErrorLimit: 3
+};
+export const MODEL_PRICING = {
+  // Anthropic direct — Claude 4.x
+  "claude-sonnet-4-6": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "claude-sonnet-4-20250514": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "claude-sonnet-4-5-20250929": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "claude-opus-4-6": {
+    inputPer1M: 5.0,
+    outputPer1M: 25.0,
+    thinkingPer1M: 25.0
+  },
+  "claude-opus-4-20250514": {
+    inputPer1M: 5.0,
+    outputPer1M: 25.0,
+    thinkingPer1M: 25.0
+  },
+  "claude-opus-4-5-20251101": {
+    inputPer1M: 5.0,
+    outputPer1M: 25.0,
+    thinkingPer1M: 25.0
+  },
+  "claude-haiku-4-20250514": {
+    inputPer1M: 1.0,
+    outputPer1M: 5.0,
+    thinkingPer1M: 5.0
+  },
+  "claude-haiku-4-5-20251001": {
+    inputPer1M: 1.0,
+    outputPer1M: 5.0,
+    thinkingPer1M: 5.0
+  },
+  // Anthropic direct — Claude 3.5
+  "claude-3-5-sonnet-20241022": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0
+  },
+  "claude-3-5-haiku-20241022": {
+    inputPer1M: 0.80,
+    outputPer1M: 4.0
+  },
+  // Bedrock — Claude 4.x
+  "anthropic.claude-sonnet-4-6": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "us.anthropic.claude-sonnet-4-20250514-v1:0": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "us.anthropic.claude-sonnet-4-5-20250929-v1:0": {
+    inputPer1M: 3.0,
+    outputPer1M: 15.0,
+    thinkingPer1M: 15.0
+  },
+  "us.anthropic.claude-haiku-4-5-20251001-v1:0": {
+    inputPer1M: 1.0,
+    outputPer1M: 5.0,
+    thinkingPer1M: 5.0
+  },
+  // Bedrock — Claude 3.5
+  "us.anthropic.claude-3-5-haiku-20241022-v1:0": {
+    inputPer1M: 0.80,
+    outputPer1M: 4.0
+  },
+  // Bedrock — Llama & Nova
+  "us.meta.llama3-1-70b-instruct-v1:0": {
+    inputPer1M: 0.72,
+    outputPer1M: 0.72
+  },
+  "us.amazon.nova-pro-v1:0": {
+    inputPer1M: 0.80,
+    outputPer1M: 3.20
+  },
+  // Gemini (thinking tokens are cheaper than output)
+  "gemini-3-pro-preview": {
+    inputPer1M: 1.25,
+    outputPer1M: 10.0,
+    thinkingPer1M: 2.50
+  },
+  "gemini-3-flash-preview": {
+    inputPer1M: 0.15,
+    outputPer1M: 0.60,
+    thinkingPer1M: 0.15
+  },
+  "gemini-2.5-pro": {
+    inputPer1M: 1.25,
+    outputPer1M: 10.0,
+    thinkingPer1M: 2.50
+  },
+  "gemini-2.5-flash": {
+    inputPer1M: 0.15,
+    outputPer1M: 0.60,
+    thinkingPer1M: 0.15
+  },
+  "gemini-2.5-flash-lite": {
+    inputPer1M: 0.075,
+    outputPer1M: 0.30,
+    thinkingPer1M: 0.075
+  },
+  "gemini-2.0-flash": {
+    inputPer1M: 0.10,
+    outputPer1M: 0.40
+  },
+  "gemini-2.0-flash-lite": {
+    inputPer1M: 0.075,
+    outputPer1M: 0.30
+  },
+  // OpenAI — GPT
+  "gpt-5": {
+    inputPer1M: 1.25,
+    outputPer1M: 10.0
+  },
+  "gpt-5-mini": {
+    inputPer1M: 0.25,
+    outputPer1M: 2.0
+  },
+  "gpt-5-nano": {
+    inputPer1M: 0.05,
+    outputPer1M: 0.40
+  },
+  "gpt-4o": {
+    inputPer1M: 2.50,
+    outputPer1M: 10.0
+  },
+  "gpt-4o-mini": {
+    inputPer1M: 0.15,
+    outputPer1M: 0.60
+  },
+  "gpt-4-turbo": {
+    inputPer1M: 10.0,
+    outputPer1M: 30.0
+  },
+  // OpenAI — reasoning models
+  "o1": {
+    inputPer1M: 15.0,
+    outputPer1M: 60.0,
+    thinkingPer1M: 60.0
+  },
+  "o1-mini": {
+    inputPer1M: 3.0,
+    outputPer1M: 12.0,
+    thinkingPer1M: 12.0
+  },
+  "o3": {
+    inputPer1M: 2.0,
+    outputPer1M: 8.0,
+    thinkingPer1M: 8.0
+  },
+  "o3-mini": {
+    inputPer1M: 1.10,
+    outputPer1M: 4.40,
+    thinkingPer1M: 4.40
+  },
+  "o4-mini": {
+    inputPer1M: 1.10,
+    outputPer1M: 4.40,
+    thinkingPer1M: 4.40
+  }
+};
+const DEFAULT_OUTPUT_TOKENS = 32000;
+const MODEL_MAX_OUTPUT_TOKENS = {
+  // Anthropic — current models (actual API-enforced limits)
+  "claude-opus-4-6": 128000,
+  // 128K
+  "claude-sonnet-4-6": 64000,
+  // 64K (API enforces 64000, not 65536)
+  "claude-haiku-4-5-20251001": 64000,
+  // 64K
+  // Anthropic — legacy models
+  "claude-sonnet-4-5-20250929": 64000,
+  // 64K (API enforces 64000)
+  "claude-opus-4-5-20251101": 64000,
+  // 64K
+  "claude-opus-4-1-20250805": 32768,
+  // 32K
+  "claude-sonnet-4-20250514": 64000,
+  // 64K
+  "claude-opus-4-20250514": 32768,
+  // 32K
+  "claude-haiku-4-20250514": 64000,
+  // 64K
+  "claude-3-5-sonnet-20241022": 8192,
+  // 8K
+  "claude-3-5-haiku-20241022": 8192 // 8K
+};
+export function getMaxOutputTokens(model, agentMax) {
+  const modelMax = MODEL_MAX_OUTPUT_TOKENS[model] ?? DEFAULT_OUTPUT_TOKENS;
+  if (agentMax) return Math.min(agentMax, modelMax);
+  return Math.min(DEFAULT_OUTPUT_TOKENS, modelMax);
+}
+//# sourceMappingURL=agent-core-types.js.map

package/dist/shared/agent-core-types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-core-types.js","names":["AGENT_DEFAULTS","maxTurns","temperature","maxTokens","maxConcurrentTools","enableDelegation","maxDurationMs","maxDurationMsWorkflow","enableModelRouting","maxMessageChars","maxHistoryChars","maxToolResultChars","subagentMaxTokens","subagentMaxTurns","subagentTemperature","maxContinuations","subagentMaxContinuations","subagentMaxRetries","teammateMaxTotalTurns","teamTimeoutMs","workerStallMs","teamReviewMaxTokens","apiTimeoutMs","toolTimeoutMs","toolResultTruncateChars","compactionTriggerTokens","loopIdenticalCallLimit","loopConsecutiveErrorLimit","loopTurnErrorLimit","loopWindow","loopSessionToolErrorLimit","loopConsecutiveFailedTurnLimit","loopFileReadLimit","loopBatchErrorLimit","MODEL_PRICING","inputPer1M","outputPer1M","thinkingPer1M","DEFAULT_OUTPUT_TOKENS","MODEL_MAX_OUTPUT_TOKENS","getMaxOutputTokens","model","agentMax","modelMax","Math","min"],"sources":["../../src/shared/agent-core-types.ts"],"sourcesContent":["/**\n * Agent Core Types — shared type definitions\n * Extracted from agent-core.ts for modularity.\n */\n\n// ============================================================================\n// TYPES\n// ============================================================================\n\nexport interface LoopCheckResult {\n  blocked: boolean;\n  reason?: string;\n}\n\nexport interface BailCheckResult {\n  shouldBail: boolean;\n  message?: string;\n}\n\nexport interface CostInfo {\n  inputTokens: number;\n  outputTokens: number;\n  costUsd: number;\n  cacheCreation: number;\n  cacheRead: number;\n}\n\n/**\n * Controls which tools the model can/must use:\n * - \"auto\": Model decides whether to use tools (default)\n * - \"any\": Model must use at least one tool\n * - \"none\": Model must not use any tools (text-only response)\n * - { type: \"tool\", name: string }: Model must use the specified tool\n */\nexport type ToolChoice = \"auto\" | \"any\" | \"none\" | { type: \"tool\"; name: string };\n\nexport interface ContextManagementConfig {\n  betas: string[];\n  config: { edits: Array<Record<string, unknown>> };\n}\n\n/** Overrides from ai_agent_config.context_config JSONB — all optional, falls back to defaults */\nexport interface ContextManagementOverrides {\n  compaction_trigger_tokens?: number;\n  compaction_total_budget?: number;\n  clear_thinking_keep?: number;\n  clear_tool_uses_trigger?: number;\n  clear_tool_uses_keep?: number;\n  session_cost_budget_usd?: number;\n}\n\nexport interface CompactionConfig {\n  triggerTokens: number;\n  totalBudget: number;\n  isNative: boolean;\n}\n\nexport interface ThinkingConfig {\n  type: \"enabled\" | \"disabled\" | \"adaptive\";\n  budget_tokens?: number;\n}\n\nexport type ErrorCategory =\n  | \"RATE_LIMIT\"\n  | \"TIMEOUT\"\n  | \"AUTH\"\n  | \"MALFORMED\"\n  | \"PROVIDER_DOWN\"\n  | \"NETWORK\"\n  | \"UNKNOWN\";\n\nexport interface LoopDetectorConfig {\n  identicalCallLimit: number;\n  consecutiveErrorLimit: number;\n  turnErrorLimit: number;\n  window: number;\n  sessionToolErrorLimit: number;\n  consecutiveFailedTurnLimit: number;\n  fileReadLimit: number;\n}\n\nexport type CallPath = \"sse\" | \"workflow\" | \"channel\";\n\n/** Context config shape from ai_agent_config.context_config JSONB */\nexport interface AgentContextConfig {\n  includeLocations?: boolean;\n  locationIds?: string[];\n  includeCustomers?: boolean;\n  customerSegments?: string[];\n  max_history_chars?: number;\n  max_tool_result_chars?: number;\n  max_message_chars?: number;\n  compaction_trigger_tokens?: number;\n  compaction_total_budget?: number;\n  clear_thinking_keep?: number;\n  clear_tool_uses_trigger?: number;\n  clear_tool_uses_keep?: number;\n  session_cost_budget_usd?: number;\n  max_concurrent_tools?: number;\n  enable_delegation?: boolean;\n  max_duration_ms?: number;\n  enable_model_routing?: boolean;\n  subagent_max_tokens?: number;\n  subagent_max_turns?: number;\n  subagent_temperature?: number;\n  subagent_max_retries?: number;\n  max_continuations?: number;\n  subagent_max_continuations?: number;\n  teammate_max_total_turns?: number;\n  team_timeout_ms?: number;\n  worker_stall_ms?: number;\n  team_review_max_tokens?: number;\n  api_timeout_ms?: number;\n  tool_timeout_ms?: number;\n  tool_result_truncate_chars?: number;\n  loop_identical_call_limit?: number;\n  loop_consecutive_error_limit?: number;\n  loop_turn_error_limit?: number;\n  loop_window?: number;\n  loop_session_tool_error_limit?: number;\n  loop_consecutive_failed_turn_limit?: number;\n  loop_file_read_limit?: number;\n  loop_batch_error_limit?: number;\n}\n\nexport interface ResolvedAgentLoopConfig {\n  maxTurns: number;\n  temperature: number;\n  maxTokens: number;\n  maxConcurrentTools: number;\n  enableDelegation: boolean;\n  maxDurationMs: number;\n  enableModelRouting: boolean;\n  maxMessageChars: number;\n  maxHistoryChars: number;\n  contextOverrides: ContextManagementOverrides;\n  subagentMaxTokens: number;\n  subagentMaxTurns: number;\n  subagentTemperature: number;\n  maxContinuations: number;\n  subagentMaxContinuations: number;\n  subagentMaxRetries: number;\n  maxToolResultChars: number;\n  teammateMaxTotalTurns: number;\n  teamTimeoutMs: number;\n  workerStallMs: number;\n  teamReviewMaxTokens: number;\n  apiTimeoutMs: number;\n  toolTimeoutMs: number;\n  toolResultTruncateChars: number;\n  loopIdenticalCallLimit: number;\n  loopConsecutiveErrorLimit: number;\n  loopTurnErrorLimit: number;\n  loopWindow: number;\n  loopSessionToolErrorLimit: number;\n  loopConsecutiveFailedTurnLimit: number;\n  loopFileReadLimit: number;\n  loopBatchErrorLimit: number;\n  defaultsUsed: string[];\n}\n\n// ============================================================================\n// CONSTANTS\n// ============================================================================\n\nexport const AGENT_DEFAULTS = {\n  maxTurns: 25,\n  temperature: 0.7,\n  maxTokens: 16384,\n  maxConcurrentTools: 4,\n  enableDelegation: true,\n  maxDurationMs: 120_000,\n  maxDurationMsWorkflow: 300_000,\n  enableModelRouting: false,\n  maxMessageChars: 100_000,\n  maxHistoryChars: 500_000,\n  maxToolResultChars: 100_000,\n  subagentMaxTokens: 16384,\n  subagentMaxTurns: 15,\n  subagentTemperature: 0.7,\n  maxContinuations: 5,\n  subagentMaxContinuations: 3,\n  subagentMaxRetries: 2,\n  teammateMaxTotalTurns: 100,\n  teamTimeoutMs: 600_000,\n  workerStallMs: 120_000,\n  teamReviewMaxTokens: 16384,\n  apiTimeoutMs: 120_000,\n  toolTimeoutMs: 120_000,\n  toolResultTruncateChars: 50_000,\n  compactionTriggerTokens: 167_000,\n  // Loop detector\n  loopIdenticalCallLimit: 4,\n  loopConsecutiveErrorLimit: 5,\n  loopTurnErrorLimit: 10,\n  loopWindow: 20,\n  loopSessionToolErrorLimit: 25,\n  loopConsecutiveFailedTurnLimit: 3,\n  loopFileReadLimit: 8,\n  loopBatchErrorLimit: 3,\n} as const;\n\nexport const MODEL_PRICING: Record<string, { inputPer1M: number; outputPer1M: number; thinkingPer1M?: number }> = {\n  // Anthropic direct — Claude 4.x\n  \"claude-sonnet-4-6\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"claude-sonnet-4-20250514\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"claude-sonnet-4-5-20250929\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"claude-opus-4-6\": { inputPer1M: 5.0, outputPer1M: 25.0, thinkingPer1M: 25.0 },\n  \"claude-opus-4-20250514\": { inputPer1M: 5.0, outputPer1M: 25.0, thinkingPer1M: 25.0 },\n  \"claude-opus-4-5-20251101\": { inputPer1M: 5.0, outputPer1M: 25.0, thinkingPer1M: 25.0 },\n  \"claude-haiku-4-20250514\": { inputPer1M: 1.0, outputPer1M: 5.0, thinkingPer1M: 5.0 },\n  \"claude-haiku-4-5-20251001\": { inputPer1M: 1.0, outputPer1M: 5.0, thinkingPer1M: 5.0 },\n  // Anthropic direct — Claude 3.5\n  \"claude-3-5-sonnet-20241022\": { inputPer1M: 3.0, outputPer1M: 15.0 },\n  \"claude-3-5-haiku-20241022\": { inputPer1M: 0.80, outputPer1M: 4.0 },\n  // Bedrock — Claude 4.x\n  \"anthropic.claude-sonnet-4-6\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"us.anthropic.claude-sonnet-4-20250514-v1:0\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"us.anthropic.claude-sonnet-4-5-20250929-v1:0\": { inputPer1M: 3.0, outputPer1M: 15.0, thinkingPer1M: 15.0 },\n  \"us.anthropic.claude-haiku-4-5-20251001-v1:0\": { inputPer1M: 1.0, outputPer1M: 5.0, thinkingPer1M: 5.0 },\n  // Bedrock — Claude 3.5\n  \"us.anthropic.claude-3-5-haiku-20241022-v1:0\": { inputPer1M: 0.80, outputPer1M: 4.0 },\n  // Bedrock — Llama & Nova\n  \"us.meta.llama3-1-70b-instruct-v1:0\": { inputPer1M: 0.72, outputPer1M: 0.72 },\n  \"us.amazon.nova-pro-v1:0\": { inputPer1M: 0.80, outputPer1M: 3.20 },\n  // Gemini (thinking tokens are cheaper than output)\n  \"gemini-3-pro-preview\": { inputPer1M: 1.25, outputPer1M: 10.0, thinkingPer1M: 2.50 },\n  \"gemini-3-flash-preview\": { inputPer1M: 0.15, outputPer1M: 0.60, thinkingPer1M: 0.15 },\n  \"gemini-2.5-pro\": { inputPer1M: 1.25, outputPer1M: 10.0, thinkingPer1M: 2.50 },\n  \"gemini-2.5-flash\": { inputPer1M: 0.15, outputPer1M: 0.60, thinkingPer1M: 0.15 },\n  \"gemini-2.5-flash-lite\": { inputPer1M: 0.075, outputPer1M: 0.30, thinkingPer1M: 0.075 },\n  \"gemini-2.0-flash\": { inputPer1M: 0.10, outputPer1M: 0.40 },\n  \"gemini-2.0-flash-lite\": { inputPer1M: 0.075, outputPer1M: 0.30 },\n  // OpenAI — GPT\n  \"gpt-5\": { inputPer1M: 1.25, outputPer1M: 10.0 },\n  \"gpt-5-mini\": { inputPer1M: 0.25, outputPer1M: 2.0 },\n  \"gpt-5-nano\": { inputPer1M: 0.05, outputPer1M: 0.40 },\n  \"gpt-4o\": { inputPer1M: 2.50, outputPer1M: 10.0 },\n  \"gpt-4o-mini\": { inputPer1M: 0.15, outputPer1M: 0.60 },\n  \"gpt-4-turbo\": { inputPer1M: 10.0, outputPer1M: 30.0 },\n  // OpenAI — reasoning models\n  \"o1\": { inputPer1M: 15.0, outputPer1M: 60.0, thinkingPer1M: 60.0 },\n  \"o1-mini\": { inputPer1M: 3.0, outputPer1M: 12.0, thinkingPer1M: 12.0 },\n  \"o3\": { inputPer1M: 2.0, outputPer1M: 8.0, thinkingPer1M: 8.0 },\n  \"o3-mini\": { inputPer1M: 1.10, outputPer1M: 4.40, thinkingPer1M: 4.40 },\n  \"o4-mini\": { inputPer1M: 1.10, outputPer1M: 4.40, thinkingPer1M: 4.40 },\n};\n\nconst DEFAULT_OUTPUT_TOKENS = 32000;\n\nconst MODEL_MAX_OUTPUT_TOKENS: Record<string, number> = {\n  // Anthropic — current models (actual API-enforced limits)\n  \"claude-opus-4-6\": 128000,           // 128K\n  \"claude-sonnet-4-6\": 64000,          // 64K (API enforces 64000, not 65536)\n  \"claude-haiku-4-5-20251001\": 64000,  // 64K\n  // Anthropic — legacy models\n  \"claude-sonnet-4-5-20250929\": 64000, // 64K (API enforces 64000)\n  \"claude-opus-4-5-20251101\": 64000,   // 64K\n  \"claude-opus-4-1-20250805\": 32768,   // 32K\n  \"claude-sonnet-4-20250514\": 64000,   // 64K\n  \"claude-opus-4-20250514\": 32768,     // 32K\n  \"claude-haiku-4-20250514\": 64000,    // 64K\n  \"claude-3-5-sonnet-20241022\": 8192,  // 8K\n  \"claude-3-5-haiku-20241022\": 8192,   // 8K\n};\n\nexport function getMaxOutputTokens(model: string, agentMax?: number): number {\n  const modelMax = MODEL_MAX_OUTPUT_TOKENS[model] ?? DEFAULT_OUTPUT_TOKENS;\n  if (agentMax) return Math.min(agentMax, modelMax);\n  return Math.min(DEFAULT_OUTPUT_TOKENS, modelMax);\n}\n"],"mappings":"AAAA;AACA;AACA;AACA;;AAEA;AACA;AACA;;AAoBA;AACA;AACA;AACA;AACA;AACA;AACA;;AAQA;;AA0CA;;AA8EA;AACA;AACA;;AAEA,OAAO,MAAMA,cAAc,GAAG;EAC5BC,QAAQ,EAAE,EAAE;EACZC,WAAW,EAAE,GAAG;EAChBC,SAAS,EAAE,KAAK;EAChBC,kBAAkB,EAAE,CAAC;EACrBC,gBAAgB,EAAE,IAAI;EACtBC,aAAa,EAAE,OAAO;EACtBC,qBAAqB,EAAE,OAAO;EAC9BC,kBAAkB,EAAE,KAAK;EACzBC,eAAe,EAAE,OAAO;EACxBC,eAAe,EAAE,OAAO;EACxBC,kBAAkB,EAAE,OAAO;EAC3BC,iBAAiB,EAAE,KAAK;EACxBC,gBAAgB,EAAE,EAAE;EACpBC,mBAAmB,EAAE,GAAG;EACxBC,gBAAgB,EAAE,CAAC;EACnBC,wBAAwB,EAAE,CAAC;EAC3BC,kBAAkB,EAAE,CAAC;EACrBC,qBAAqB,EAAE,GAAG;EAC1BC,aAAa,EAAE,OAAO;EACtBC,aAAa,EAAE,OAAO;EACtBC,mBAAmB,EAAE,KAAK;EAC1BC,YAAY,EAAE,OAAO;EACrBC,aAAa,EAAE,OAAO;EACtBC,uBAAuB,EAAE,MAAM;EAC/BC,uBAAuB,EAAE,OAAO;EAChC;EACAC,sBAAsB,EAAE,CAAC;EACzBC,yBAAyB,EAAE,CAAC;EAC5BC,kBAAkB,EAAE,EAAE;EACtBC,UAAU,EAAE,EAAE;EACdC,yBAAyB,EAAE,EAAE;EAC7BC,8BAA8B,EAAE,CAAC;EACjCC,iBAAiB,EAAE,CAAC;EACpBC,mBAAmB,EAAE;AACvB,CAAU;AAEV,OAAO,MAAMC,aAAkG,GAAG;EAChH;EACA,mBAAmB,EAAE;IAAEC,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAChF,0BAA0B,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACvF,4BAA4B,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACzF,iBAAiB,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAC9E,wBAAwB,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACrF,0BAA0B,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACvF,yBAAyB,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,GAAG;IAAEC,aAAa,EAAE;EAAI,CAAC;EACpF,2BAA2B,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,GAAG;IAAEC,aAAa,EAAE;EAAI,CAAC;EACtF;EACA,4BAA4B,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE;EAAK,CAAC;EACpE,2BAA2B,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAI,CAAC;EACnE;EACA,6BAA6B,EAAE;IAAED,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAC1F,4CAA4C,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACzG,8CAA8C,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAC3G,6CAA6C,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,GAAG;IAAEC,aAAa,EAAE;EAAI,CAAC;EACxG;EACA,6CAA6C,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAI,CAAC;EACrF;EACA,oCAAoC,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EAC7E,yBAAyB,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EAClE;EACA,sBAAsB,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACpF,wBAAwB,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACtF,gBAAgB,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAC9E,kBAAkB,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAChF,uBAAuB,EAAE;IAAEF,UAAU,EAAE,KAAK;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAM,CAAC;EACvF,kBAAkB,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EAC3D,uBAAuB,EAAE;IAAED,UAAU,EAAE,KAAK;IAAEC,WAAW,EAAE;EAAK,CAAC;EACjE;EACA,OAAO,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EAChD,YAAY,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAI,CAAC;EACpD,YAAY,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EACrD,QAAQ,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EACjD,aAAa,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EACtD,aAAa,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE;EAAK,CAAC;EACtD;EACA,IAAI,EAAE;IAAED,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EAClE,SAAS,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACtE,IAAI,EAAE;IAAEF,UAAU,EAAE,GAAG;IAAEC,WAAW,EAAE,GAAG;IAAEC,aAAa,EAAE;EAAI,CAAC;EAC/D,SAAS,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK,CAAC;EACvE,SAAS,EAAE;IAAEF,UAAU,EAAE,IAAI;IAAEC,WAAW,EAAE,IAAI;IAAEC,aAAa,EAAE;EAAK;AACxE,CAAC;AAED,MAAMC,qBAAqB,GAAG,KAAK;AAEnC,MAAMC,uBAA+C,GAAG;EACtD;EACA,iBAAiB,EAAE,MAAM;EAAY;EACrC,mBAAmB,EAAE,KAAK;EAAW;EACrC,2BAA2B,EAAE,KAAK;EAAG;EACrC;EACA,4BAA4B,EAAE,KAAK;EAAE;EACrC,0BAA0B,EAAE,KAAK;EAAI;EACrC,0BAA0B,EAAE,KAAK;EAAI;EACrC,0BAA0B,EAAE,KAAK;EAAI;EACrC,wBAAwB,EAAE,KAAK;EAAM;EACrC,yBAAyB,EAAE,KAAK;EAAK;EACrC,4BAA4B,EAAE,IAAI;EAAG;EACrC,2BAA2B,EAAE,IAAI,CAAI;AACvC,CAAC;AAED,OAAO,SAASC,kBAAkBA,CAACC,KAAa,EAAEC,QAAiB,EAAU;EAC3E,MAAMC,QAAQ,GAAGJ,uBAAuB,CAACE,KAAK,CAAC,IAAIH,qBAAqB;EACxE,IAAII,QAAQ,EAAE,OAAOE,IAAI,CAACC,GAAG,CAACH,QAAQ,EAAEC,QAAQ,CAAC;EACjD,OAAOC,IAAI,CAACC,GAAG,CAACP,qBAAqB,EAAEK,QAAQ,CAAC;AAClD","ignoreList":[]}

package/dist/shared/agent-core.d.ts

@@ -4,6 +4,7 @@
  * Pure TypeScript, no runtime-specific APIs (no Deno.env, no process.env, no fs).
  * Both the CLI (Node.js) and server (Fly container) import from here.
  */
+import type { CompactableMessage } from "./compaction.js";
 export interface LoopCheckResult {
     blocked: boolean;
     reason?: string;
@@ -61,13 +62,12 @@ export interface ContextManagementOverrides {
     clear_tool_uses_keep?: number;
     session_cost_budget_usd?: number;
 }
-/** Legacy compaction trigger — used by getCompactionConfig() for non-Anthropic providers.
- *  For Anthropic models, getContextManagement() with DB overrides takes precedence. */
-export declare const COMPACTION_TRIGGER_TOKENS = 120000;
-/** Max cumulative tokens before forcing wrap-up (prevents runaway compaction cost) */
+/** Legacy compaction trigger — used by getCompactionConfig() for non-Anthropic providers. */
+export declare const COMPACTION_TRIGGER_TOKENS = 167000;
+/** Max cumulative tokens before forcing wrap-up */
 export declare const COMPACTION_TOTAL_BUDGET = 2000000;
-/** Default session cost budget in USD — hard cap to prevent runaway spending */
-export declare const DEFAULT_SESSION_COST_BUDGET_USD = 5;
+/** Session cost safety cap — DB context_config.session_cost_budget_usd overrides */
+export declare const DEFAULT_SESSION_COST_BUDGET_USD = 25;
 /**
  * Provider-aware compaction configuration.
  * - Anthropic/Bedrock: native server-side compaction via compact_20260112
@@ -94,12 +94,21 @@ export declare function getMaxOutputTokens(model: string, agentMax?: number): nu
  * - Turn boundary (second-to-last message)
  * System prompt caching is handled by the caller.
  */
-export declare function addPromptCaching(tools: Array<Record<string, unknown>>, messages: Array<Record<string, unknown>>): {
-    tools: Array<Record<string, unknown>>;
-    messages: Array<Record<string, unknown>>;
+export declare function addPromptCaching<T extends object = Record<string, unknown>, M extends CompactableMessage = CompactableMessage>(tools: T[], messages: M[]): {
+    tools: T[];
+    messages: M[];
 };
 /** djb2 string hash — fast, deterministic, no dependencies */
 export declare function djb2Hash(str: string): string;
+export interface LoopDetectorConfig {
+    identicalCallLimit: number;
+    consecutiveErrorLimit: number;
+    turnErrorLimit: number;
+    window: number;
+    sessionToolErrorLimit: number;
+    consecutiveFailedTurnLimit: number;
+    fileReadLimit: number;
+}
 export declare class LoopDetector {
     private history;
     private consecutiveErrors;
@@ -111,13 +120,11 @@ export declare class LoopDetector {
     private totalSessionErrors;
     /** Tracks how many times the same file path has been read this session */
     private fileReadCounts;
-    static IDENTICAL_CALL_LIMIT: number;
-    static CONSECUTIVE_ERROR_LIMIT: number;
-    static TURN_ERROR_LIMIT: number;
-    static WINDOW: number;
-    static SESSION_TOOL_ERROR_LIMIT: number;
-    static CONSECUTIVE_FAILED_TURN_LIMIT: number;
-    static FILE_READ_LIMIT: number;
+    /** Resolved config — accepts DB overrides, falls back to AGENT_DEFAULTS */
+    readonly config: LoopDetectorConfig;
+    constructor(config?: Partial<LoopDetectorConfig>);
+    /** Create a LoopDetector from a ResolvedAgentLoopConfig */
+    static fromResolved(resolved: ResolvedAgentLoopConfig): LoopDetector;
     /** Get the error-tracking key for a tool call. Tools with an `action` param
      *  are tracked per-action so e.g. voice/speak failing won't block voice/music_compose. */
     private errorKey;
@@ -127,7 +134,14 @@ export declare class LoopDetector {
      * Blocks re-reading the same path more than FILE_READ_LIMIT times per session.
      */
     trackRead(path: string): LoopCheckResult;
-    recordResult(name: string, success: boolean, input?: Record<string, unknown>): void;
+    /**
+     * Record a tool execution result.
+     * @param transient - If true, the failure is transient (timeout, rate limit)
+     *   and should NOT be added to failedStrategies. Transient failures are
+     *   retryable with the exact same inputs — e.g., a task_output poll that
+     *   timed out because the task is still running.
+     */
+    recordResult(name: string, success: boolean, input?: Record<string, unknown>, transient?: boolean): void;
     endTurn(): BailCheckResult;
     resetTurn(): void;
     reset(): void;
@@ -180,25 +194,63 @@ export declare function isRetryableError(err: unknown): boolean;
  * - sonnet: kept for plan, demoted to haiku for others
  * - default/undefined: haiku
  */
-export declare function demoteSubagentModel(requested: string | undefined, agentType?: string): "haiku" | "sonnet";
+/**
+ * Resolve subagent model based on agent type and request.
+ *
+ * Claude Code behavior: general-purpose and plan subagents inherit the parent
+ * model (they do real work / need full reasoning). Only explore and research
+ * are demoted to haiku (fast, focused tasks).
+ */
+export declare function demoteSubagentModel(requested: string | undefined, agentType?: string): "haiku" | "sonnet" | "opus";
 export type CallPath = "sse" | "workflow" | "channel";
-/** Documented defaults — replaces scattered magic numbers across call paths */
+/**
+ * EMERGENCY FALLBACK DEFAULTS — used ONLY when ai_agent_config DB is unreachable.
+ *
+ * The SINGLE SOURCE OF TRUTH is the ai_agent_config table (Supabase), specifically
+ * the row for Whale Code (id: 72d7aaa8-2e8b-48a0-a38d-f332f69600bb).
+ * All behavioral knobs are stored in `context_config` JSONB.
+ *
+ * These values MUST mirror what's in the DB. If you change a value, update the DB
+ * first (via Supabase dashboard or migration), then mirror it here.
+ *
+ * DB → resolveAgentLoopConfig() → code reads resolved values.
+ * AGENT_DEFAULTS → code reads ONLY if DB fetch returns null.
+ */
 export declare const AGENT_DEFAULTS: {
-    readonly maxTurns: 10;
-    readonly temperature: 0.7;
+    readonly maxTurns: 0;
+    readonly temperature: 1;
     readonly maxConcurrentTools: 7;
     readonly enableDelegation: true;
     readonly maxDurationMs: 900000;
     readonly maxDurationMsWorkflow: 120000;
     readonly enableModelRouting: true;
-    readonly subagentMaxTokens: 8192;
-    readonly subagentMaxTurns: 6;
-    readonly subagentTemperature: 0.3;
+    readonly subagentMaxTokens: 32000;
+    readonly subagentMaxTurns: 0;
+    readonly subagentTemperature: 1;
+    readonly subagentMaxRetries: 2;
+    readonly maxContinuations: 10;
+    readonly subagentMaxContinuations: 5;
     readonly maxMessageChars: 100000;
     readonly maxHistoryChars: 400000;
-    readonly compactionTriggerTokens: 150000;
+    readonly maxToolResultChars: 2048;
+    readonly compactionTriggerTokens: 167000;
     readonly compactionTotalBudget: 2000000;
-    readonly sessionCostBudgetUsd: 5;
+    readonly sessionCostBudgetUsd: 25;
+    readonly teammateMaxTotalTurns: 0;
+    readonly teamTimeoutMs: 600000;
+    readonly workerStallMs: 90000;
+    readonly teamReviewMaxTokens: 1024;
+    readonly apiTimeoutMs: 90000;
+    readonly toolTimeoutMs: 60000;
+    readonly toolResultTruncateChars: 30000;
+    readonly loopIdenticalCallLimit: 4;
+    readonly loopConsecutiveErrorLimit: 3;
+    readonly loopTurnErrorLimit: 5;
+    readonly loopWindow: 20;
+    readonly loopSessionToolErrorLimit: 10;
+    readonly loopConsecutiveFailedTurnLimit: 3;
+    readonly loopFileReadLimit: 3;
+    readonly loopBatchErrorLimit: 3;
 };
 /** Context config shape from ai_agent_config.context_config JSONB */
 export interface AgentContextConfig {
@@ -222,6 +274,24 @@ export interface AgentContextConfig {
     subagent_max_tokens?: number;
     subagent_max_turns?: number;
     subagent_temperature?: number;
+    subagent_max_retries?: number;
+    max_continuations?: number;
+    subagent_max_continuations?: number;
+    teammate_max_total_turns?: number;
+    team_timeout_ms?: number;
+    worker_stall_ms?: number;
+    team_review_max_tokens?: number;
+    api_timeout_ms?: number;
+    tool_timeout_ms?: number;
+    tool_result_truncate_chars?: number;
+    loop_identical_call_limit?: number;
+    loop_consecutive_error_limit?: number;
+    loop_turn_error_limit?: number;
+    loop_window?: number;
+    loop_session_tool_error_limit?: number;
+    loop_consecutive_failed_turn_limit?: number;
+    loop_file_read_limit?: number;
+    loop_batch_error_limit?: number;
 }
 export interface ResolvedAgentLoopConfig {
     maxTurns: number;
@@ -237,6 +307,25 @@ export interface ResolvedAgentLoopConfig {
     subagentMaxTokens: number;
     subagentMaxTurns: number;
     subagentTemperature: number;
+    maxContinuations: number;
+    subagentMaxContinuations: number;
+    subagentMaxRetries: number;
+    maxToolResultChars: number;
+    teammateMaxTotalTurns: number;
+    teamTimeoutMs: number;
+    workerStallMs: number;
+    teamReviewMaxTokens: number;
+    apiTimeoutMs: number;
+    toolTimeoutMs: number;
+    toolResultTruncateChars: number;
+    loopIdenticalCallLimit: number;
+    loopConsecutiveErrorLimit: number;
+    loopTurnErrorLimit: number;
+    loopWindow: number;
+    loopSessionToolErrorLimit: number;
+    loopConsecutiveFailedTurnLimit: number;
+    loopFileReadLimit: number;
+    loopBatchErrorLimit: number;
     /** Which fields fell back to AGENT_DEFAULTS (transparency for logging) */
     defaultsUsed: string[];
 }