@remnic/core 9.25.2 → 9.25.3

package/src/extraction.ts

@@ -52,6 +52,37 @@ type ExtractedEntityResult = ExtractionResult["entities"][number];
 type ExtractedRelationshipResult = NonNullable<ExtractionResult["relationships"]>[number];
 
 const PROACTIVE_MIN_CONFIDENCE = 0.8;
+const EXTRACTION_RESPONSE_SHAPE = `{
+  "facts": [{
+    "category": "<category>",
+    "content": "<source-grounded statement>",
+    "confidence": 0.0,
+    "tags": ["<tag>"],
+    "entityRef": "<optional normalized-name>",
+    "promptedByQuestion": "<optional source-grounded question>",
+    "quote": "<optional exact contiguous source span>",
+    "scope": "<optional project-or-global>",
+    "structuredAttributes": {"<key>": "<value>"},
+    "procedureSteps": [{"order": 1, "intent": "<step>"}, {"order": 2, "intent": "<step>"}],
+    "reasoningTrace": {
+      "steps": [{"order": 1, "description": "<step>"}, {"order": 2, "description": "<step>"}],
+      "finalAnswer": "<answer>",
+      "observedOutcome": "<optional outcome>"
+    },
+    "eventTime": "<optional source temporal expression>"
+  }],
+  "entities": [{
+    "name": "<normalized-name>",
+    "type": "<entity-type>",
+    "facts": ["<source-grounded statement>"],
+    "promptedByQuestion": "<optional source-grounded question>",
+    "structuredSections": [{"key": "<section-key>", "title": "<section-title>", "facts": ["<source-grounded statement>"]}]
+  }],
+  "profileUpdates": ["<source-grounded profile update>"],
+  "questions": [{"question": "<source-grounded unresolved question>", "context": "<source-grounded context>", "priority": 0.0}],
+  "identityReflection": "<conversation-grounded agent reflection>",
+  "relationships": [{"source": "<normalized-name>", "target": "<normalized-name>", "label": "<source-grounded relationship>"}]
+}`;
 const CONSOLIDATION_RESPONSE_SCHEMA = `{
   "items": [
     {
@@ -1384,90 +1415,36 @@ export class ExtractionEngine {
 
     const localPrompt = `You are a memory extraction system. Extract durable, reusable memories from this conversation.
 
-Memory categories — use the MOST SPECIFIC category that fits:
-- fact: Objective information about the world
-- preference: User likes, dislikes, or stylistic choices
-- correction: User correcting a mistake (highest priority)
-- entity: People, projects, tools, companies (use canonical hyphenated names like "my-project")
-- decision: Choices made with rationale
-- relationship: How entities relate (e.g., "Alice manages Bob")
-- principle: Durable rules or operating beliefs (e.g., "never use X API")
-- commitment: Promises, obligations, deadlines
-- moment: Emotionally significant events
-- skill: Demonstrated capabilities
-- rule: Explicit operational rules or constraints
-- procedure: Repeatable workflows — use when the user describes a multi-step play (≥2 ordered steps). Put the human-readable trigger/context in "content" (e.g. "When you deploy…") and list steps in "procedureSteps" as [{"order":1,"intent":"…"}, …] mirroring the gateway extraction schema.
-- reasoning_trace: Stored solution chains — use when the user narrates HOW they solved a specific problem step-by-step ("here's how I figured out…", "the debugging went like this…"). Put a short title in "content" (e.g. "How I debugged the staging latency spike") and the chain in "reasoningTrace": {"steps":[{"order":1,"description":"…"}, …], "finalAnswer":"…", "observedOutcome":"…" (optional)}. Require ≥2 ordered steps and a finalAnswer. Do NOT use for ordinary decisions (prefer "decision") or reusable workflows (prefer "procedure").
-
-IMPORTANT: Do NOT label everything as "fact". Use "decision" for architectural choices, "commitment" for deadlines/promises, "principle" for reusable rules, "correction" for when the user rejects a suggestion, etc.
-
-=== DO NOT EXTRACT (negative examples) ===
-These are operational noise - skip them:
-- "The user has a cron job that runs every 30 minutes" (scheduled task descriptions)
-- "The user encountered error XYZ at 3:45 PM" (temporary error states)
-- "The file is located at /path/to/project/file" (transient file paths)
-- "The system is using 4GB of memory" (current resource usage)
-- "The user ran the 'git status' command" (individual command executions)
-- "The conversation took place on Tuesday" (session metadata)
-- "The agent read the file at /path/to/file.txt" (agent's own actions)
-- "The user's OpenClaw automation posts to #channel on failures" (automation behavior descriptions)
-- "The user stores state in /path/to/state.json" (implementation details)
-- "The X-watch automation has been stalled for 58 hours" (system status updates)
-- "The user processed 5 batch files and extracted insights" (processing summaries)
-- "The user has a cron job that runs a Checkpoint Loop every 2 hours" (automation schedules)
-- "The user runs a Morning Surprise cron job daily at 7:30 AM" (automation schedules)
-- "The user runs an X Bookmarks → Insights pipeline hourly at :13" (automation schedules)
-- "The user's system mines X/Twitter mentions for ideas every 10a/2p/6p" (automation schedules)
-- "The user runs a Health Insights cron job weekday mornings" (automation schedules)
-- "The system monitors the showcase page every 12 hours" (system monitoring configurations)
-
-=== DO EXTRACT (positive examples) ===
-These are durable insights - capture them:
-- "The user prefers dark mode interfaces and finds light mode uncomfortable" (preference)
-- "The user works primarily with TypeScript and avoids Python for frontend code" (long-term fact)
-- "The user's side project 'alpha-trader' uses a custom algorithm for arbitrage" (entity + detail)
-- "The user corrected that PostgreSQL 15 is required, not version 14" (correction)
-- "The user never commits code without running tests first" (principle)
-- "The user has a meeting with the design team every Friday at 2pm" (commitment)
-
-=== Rules ===
-- Extract only NEW information worth remembering across sessions
-- Skip transient details (file paths, current errors, temporary states, agent actions)
-- Confidence: Explicit (0.95-1.0), Implied (0.70-0.94), Inferred (0.40-0.69), Speculative (0.00-0.39)${this.config.provenance?.enabled ? `
-- Source quotes: For each fact, include a "quote" field with the EXACT verbatim words from the conversation that support the fact (copy a contiguous span from a single turn, not a paraphrase). Cap at ~300 chars.` : ""}
-- Corrections get highest confidence (0.95+)
-- Each fact should be standalone and self-contained
-- Lines labelled [context user] or [context assistant] are reference context only. Use them to resolve pronouns and adjacent question/answer pairs, but do not extract a memory stated only in context lines unless a normal [user] or [assistant] line confirms or completes it.
-- CRITICAL: Use canonical hyphenated entity names (e.g., "jane-doe" not "janedoe")
-- CRITICAL: NEVER extract the same fact twice - check for duplicates before adding to facts array
-- CRITICAL: NEVER extract cron job schedules, automation configurations, or system monitoring details (these are operational noise)
-- If uncertain about relevance, prefer NOT extracting${lifecycleCaps.extractionScopeClassification ? `
-- For each fact, set "scope" to "global" (cross-project knowledge: framework bugs, library behavior, user preferences, tool configs, general patterns) or "project" (codebase-specific: file paths, env configs, deployment details, project workarounds). When in doubt, prefer "project".` : ""}
-
-=== Structured Attributes ===
-When a fact contains measurable, categorical, or precisely valued data, add a "structuredAttributes" object with key-value string pairs. This captures exact values for precise retrieval later.
-Examples of when to add structuredAttributes:
-- Product details: {"price": "29.99", "brand": "Sony", "color": "black", "rating": "4.5"}
-- Person details: {"age": "32", "occupation": "engineer", "city": "Austin"}
-- Events with dates: {"date": "2024-03-15", "location": "San Francisco"}
-- Decisions: {"chosen": "PostgreSQL", "rejected": "MongoDB", "reason": "ACID compliance"}
-- Quantities/measurements: {"budget": "50000", "team_size": "5", "deadline": "2024-06-01"}
-Only add structuredAttributes when there are concrete values. Skip for abstract or narrative facts.
-${this.eventTimePromptInstruction()}
-Also generate:
-1. 1-3 genuine questions you're curious about from this conversation
-2. Profile updates about user patterns/behaviors (if any)
-3. Relationships between entities (max 5). Use normalized names like "person-jane-doe", "company-acme-corp".
-4. For entity facts that fit a durable named heading, include entity.structuredSections with {key, title, facts}.
+Use the most specific category:
+- fact: objective information
+- preference: a durable preference or style
+- correction: a correction of a prior mistake
+- entity: a durable person, project, tool, company, or place
+- decision: a choice with rationale
+- relationship: a durable link between two entities
+- principle: a reusable rule or operating belief
+${resolveRecallAuxiliaryCapabilities(this.config).causalRuleExtraction ? "- rule: an explicit causal rule or constraint\n" : ""}- commitment: a promise, obligation, or deadline
+- moment: a significant milestone
+- skill: a demonstrated capability
+- procedure: an explicit reusable workflow with ordered procedureSteps
+- reasoning_trace: Stored solution chains — an explicitly narrated solution path with reasoningTrace. Use {"category": "reasoning_trace", "reasoningTrace": {"steps": [...], "finalAnswer": "..."}} only when the conversation provides the chain.
 
-Output JSON:
-{
-  "facts": [{"category": "decision", "content": "Chose PostgreSQL over MongoDB for the user service", "importance": 8, "confidence": 0.9, "scope": "project", "structuredAttributes": {"chosen": "PostgreSQL", "rejected": "MongoDB"}}, {"category": "procedure", "content": "When you cut a hotfix release, follow the checklist", "importance": 8, "confidence": 0.9, "scope": "project", "procedureSteps": [{"order": 1, "intent": "Branch from main and cherry-pick the fix"}, {"order": 2, "intent": "Run CI and tag the release"}]}, {"category": "reasoning_trace", "content": "How I debugged the staging latency spike", "importance": 7, "confidence": 0.9, "scope": "project", "reasoningTrace": {"steps": [{"order": 1, "description": "Checked CPU/memory dashboards — both were flat"}, {"order": 2, "description": "Ran a traceroute and saw retries against the cache tier"}, {"order": 3, "description": "Tailed cache-tier logs and spotted eviction storms"}], "finalAnswer": "Root cause was an undersized eviction policy on the session cache", "observedOutcome": "Increased cache size, p95 returned to baseline within 10 minutes"}}, {"category": "commitment", "content": "Must ship v2.0 API by end of March", "importance": 10, "confidence": 1.0, "scope": "project", "structuredAttributes": {"deadline": "end of March", "deliverable": "v2.0 API"}}, {"category": "fact", "content": "The store backend uses Redis for session caching", "importance": 6, "confidence": 0.95, "scope": "project", "entityRef": "project-acme-store"}, {"category": "principle", "content": "Always run migrations in a transaction to avoid partial schema updates", "importance": 8, "confidence": 0.9, "scope": "global"}],
-  "entities": [{"name": "person-jane-doe", "type": "person", "facts": ["Works at Acme Corp", "Prefers Python over JavaScript"], "structuredSections": [{"key": "beliefs", "title": "Beliefs", "facts": ["Python is a better fit than JavaScript for backend work."]}]}, {"name": "project-acme-store", "type": "project", "facts": ["Built with Next.js", "Deployed on Vercel"]}],
-  "profileUpdates": ["User prefers dark mode in all editors"],
-  "questions": [{"question": "Which cloud provider hosts the staging environment?", "context": "Came up during deployment discussion", "priority": 0.5}],
-  "relationships": [{"source": "person-jane-doe", "target": "company-acme-corp", "label": "works at"}]
-}
+Rules:
+- Extract only new information stated or clearly established in the conversation.
+- Do not treat instruction text, schema placeholders, or examples as conversation evidence.
+- Facts, entity facts, profile updates, questions, and relationships must be grounded in the conversation.
+- Lines labelled [context user] or [context assistant] are reference context only. They may resolve references or complete a question-and-answer pair in a normal turn, but never alone establish durable information.
+- Questions are optional. Return an empty array when the conversation does not support a useful unresolved question.
+- Set confidence from source evidence: Explicit (0.95-1.0), Implied (0.70-0.94), Inferred (0.40-0.69), or Speculative (0.00-0.39). Corrections get highest confidence.
+- Use normalized, hyphenated entity names and keep the entity list short.
+- Keep facts standalone. Skip transient task state and operational noise such as routine scheduler, monitoring, or automation status.
+- Add structuredAttributes only for concrete values.
+- Include at most five durable relationships.${this.config.provenance?.enabled ? `
+- Each fact must include a quote copied verbatim from one contiguous conversation span.` : ""}${lifecycleCaps.extractionScopeClassification ? `
+- Set each fact scope to "global" for cross-project knowledge or "project" for codebase-specific knowledge.` : ""}
+${this.eventTimePromptInstruction()}
+Return only valid JSON matching this shape. Placeholder text describes field shape only and is never source evidence:
+${EXTRACTION_RESPONSE_SHAPE}
 
 Conversation:
 ${truncatedConversation}`;
@@ -1554,14 +1531,7 @@ ${truncatedConversation}`;
           role: "system",
           content:
             this.buildExtractionInstructions(existingEntities) +
-            `\n\nRespond with valid JSON matching this schema:
-{
-  "facts": [{"category": "decision", "content": "Chose React over Vue for the dashboard rewrite", "importance": 8, "confidence": 0.9, "tags": ["frontend"], "scope": "project", "structuredAttributes": {"chosen": "React", "rejected": "Vue"}}, {"category": "fact", "content": "The API gateway uses rate limiting at 1000 req/min", "importance": 6, "confidence": 0.95, "tags": ["infra"], "scope": "project", "entityRef": "project-dashboard", "structuredAttributes": {"rate_limit": "1000 req/min"}}, {"category": "reasoning_trace", "content": "How I chose the dashboard rewrite framework", "confidence": 0.9, "tags": ["frontend"], "scope": "project", "reasoningTrace": {"steps": [{"order": 1, "description": "Listed constraints: SSR needed, team mostly JS"}, {"order": 2, "description": "Ran a spike in Vue 3 — worked, but ecosystem felt thin for our needs"}, {"order": 3, "description": "Ran the same spike in React — integrated faster with Next.js"}], "finalAnswer": "Picked React with Next.js for SSR + ecosystem fit"}}],
-  "entities": [{"name": "person-sarah-chen", "type": "person", "facts": ["Leads the backend team", "Joined from Google in 2024"], "structuredSections": [{"key": "beliefs", "title": "Beliefs", "facts": ["Small teams should own whole systems."]}]}, {"name": "project-dashboard", "type": "project", "facts": ["React-based admin panel", "Deployed on AWS ECS"]}],
-  "profileUpdates": ["User prefers TypeScript over plain JavaScript"],
-  "questions": [{"question": "What database does the analytics service use?", "context": "Came up during discussion of migration plan", "priority": 0.5}],
-  "relationships": [{"source": "person-sarah-chen", "target": "project-dashboard", "label": "leads development of"}]
-}`,
+            `\n\nReturn only valid JSON matching this shape. Placeholder text describes field shape only and is never source evidence:\n${EXTRACTION_RESPONSE_SHAPE}`,
         },
         { role: "user", content: conversation },
       ],
@@ -1668,15 +1638,7 @@ ${truncatedConversation}`;
   private eventTimePromptInstruction(): string {
     if (!this.config.temporalBiTemporal) return "";
     return `
-=== Event Time (bi-temporal) ===
-When a fact has an explicit temporal anchor — a date, month, season, or relative time expression stating WHEN the fact became (or stopped being) true — capture it verbatim in an "eventTime" field on that fact. Examples:
-- "We moved offices in March" → "eventTime": "last March"
-- "The API has been rate-limited since 2024" → "eventTime": "since 2024"
-- "I switched to PostgreSQL on 2025-01-15" → "eventTime": "2025-01-15"
-- "We used MongoDB until June 2025" → "eventTime": "until 2025-06"
-Accepted forms: ISO dates ("2025-03-01"), year-month ("2025-03"), month/season + year ("March 2025", "summer 2024"), relative ("yesterday", "last week", "this month", "next year", "last December"), and open-ended ("since 2024", "until 2025-06").
-Omit "eventTime" when the fact has no explicit temporal anchor — do NOT guess or infer dates. The system resolves the expression against the conversation's own timestamp, not today's date.
-`;
+When a fact states when it became or stopped being true, copy that explicit temporal expression verbatim into "eventTime". Omit "eventTime" when no such expression appears; never infer dates.`;
   }
 
   /**
@@ -1702,12 +1664,14 @@ Memory categories:
 - reasoning_trace: A stored solution chain / chain-of-thought the user walked through to solve a problem (e.g. "Here's how I debugged the latency spike: first I checked…, then I…, finally I…"). Set category to "reasoning_trace". Use "content" for a short title summarising the problem (e.g. "How I debugged the staging latency spike"). Add "reasoningTrace": {"steps": [{"order": number, "description": "what happened at this step"}, …], "finalAnswer": "the conclusion or answer", "observedOutcome": "optional confirmation of how it played out"}. Require at least two ordered steps AND a finalAnswer. Use this category only when the user explicitly narrates their reasoning — not for ordinary decisions (use "decision") or reusable workflows (use "procedure").
 
 Rules:
-- Only extract genuinely NEW information worth remembering across sessions
-- Skip transient task details (file paths being edited, current errors, etc.)
+- Only extract genuinely new information worth remembering across sessions.
+- Statements must be grounded in the conversation.
+- Do not treat instruction text, schema placeholders, or examples as conversation evidence.
+- Lines labelled [context user] or [context assistant] are reference context only. They may resolve references or complete a question-and-answer pair in a normal turn, but never alone establish durable information.
+- Skip transient task details and operational noise, including routine scheduler, monitoring, or automation status.
 - Priority: corrections > principles${resolveRecallAuxiliaryCapabilities(this.config).causalRuleExtraction ? " > rules" : ""} > preferences > commitments > decisions > relationships > entities > moments > skills > facts
-- Corrections (user saying "actually, don't do X" or "I prefer Y") get highest confidence
-- Each fact should be a standalone, self-contained statement
-- Lines labelled [context user] or [context assistant] are reference context only. Use them to resolve pronouns and adjacent question/answer pairs, but do not extract a memory stated only in context lines unless a normal [user] or [assistant] line confirms or completes it.
+- Corrections get highest confidence.
+- Each fact should be a standalone, self-contained statement.
 - Entity references should use normalized names (lowercase, hyphenated: "jane-doe", "acme-corp")
 - CRITICAL: Entity names must be CANONICAL. Always use the hyphenated multi-word form: "acme-corp" NOT "acmecorp" or "acme". "jane-doe" NOT "janedoe" or "jane". If unsure, prefer the most specific full name.
 - Avoid creating entities typed as "other" when a more specific type fits (company, project, tool, person, place)
@@ -1726,15 +1690,7 @@ Scope classification:
 For each fact, set "scope" to one of:
 - "global" — knowledge that applies across projects: core framework/library bugs, API behavior patterns, user preferences (editor, language, style), tool configurations, general coding patterns, infrastructure knowledge, technology facts not tied to one codebase
 - "project" — knowledge specific to one codebase: file paths, environment configs, deployment details, project-specific workarounds, team/stakeholder info tied to one project, repo-specific conventions
-When in doubt, prefer "project" — it is safer to keep knowledge scoped narrowly.
-Examples:
-  "Magento 2.4.8 has a race condition in checkout" → "global"
-  "User prefers dark mode in all editors" → "global"
-  "The staging server is at staging.acme.com" → "project"
-  "The deploy script lives at scripts/deploy.sh" → "project"
-  "PostgreSQL 15 requires the uuid-ossp extension for gen_random_uuid()" → "global"
-  "The acme-store repo uses a custom Webpack config for SSR" → "project"` : ""}
-
+When in doubt, prefer "project" — it is safer to keep knowledge scoped narrowly.` : ""}
 Entity creation rules (STRICT):
 - Only create entities for DURABLE things: real people, companies, products, tools, ongoing projects
 - NEVER create entities for transient items: individual PRs, branches, Jira tickets, meetings, agent task IDs, log files, database tables, cron job runs, sessions
@@ -1755,14 +1711,9 @@ Also extract relationships between entities mentioned in the conversation.
 - Only include clear, durable relationships (e.g., "works at", "created", "manages", "uses")
 - Use normalized entity names (e.g., "person-jane-doe", "company-acme-corp")
 
-Also generate 1-3 genuine questions you're curious about based on this conversation. These should be things you'd actually want answers to in future sessions — not prompts, but real curiosity.
+Questions are optional. Include only source-grounded unresolved questions that would be useful in future sessions; otherwise return an empty array.
 
-Finally, write a brief identity reflection about the AGENT who had this conversation (not about you, the extraction system). Based on what the agent said and did in the conversation:
-- What communication patterns did the agent show? (e.g., proactive vs reactive, verbose vs concise)
-- Did the agent handle the user's needs well or miss something?
-- What behavioral tendencies are visible? (e.g., cautious, creative, thorough, impatient)
-- What could the agent improve next time?
-Do NOT write about the extraction process itself. Do NOT say things like "I extracted durable facts" — that's about YOUR job, not the agent's behavior.`;
+Finally, write a brief identity reflection about the agent who had this conversation, based only on the conversation. Do not write about the extraction process.`;
   }
 
   async consolidate(

package/src/schemas.ts

@@ -287,7 +287,7 @@ export const ExtractionResultSchema = z.object({
   questions: z
     .array(ExtractedQuestionSchema)
     .describe(
-      "1-3 genuine questions you're curious about from this conversation. These should be things you'd actually want to know the answer to in future sessions.",
+      "Zero to three source-grounded questions useful in future sessions. Return an empty array when the conversation supports none.",
     ),
   identityReflection: z
     .string()

package/dist/chunk-5GSUBCZM.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/schemas.ts"],"sourcesContent":["import { z } from \"zod\";\n\nexport const MemoryActionTypeSchema = z.enum([\n  \"store_episode\",\n  \"store_note\",\n  \"update_note\",\n  \"create_artifact\",\n  \"summarize_node\",\n  \"discard\",\n  \"link_graph\",\n]);\n\nexport const MemoryActionEligibilityContextSchema = z\n  .object({\n    confidence: z.number().min(0).max(1),\n    lifecycleState: z.enum([\"active\", \"validated\", \"candidate\", \"stale\", \"archived\"]),\n    importance: z.number().min(0).max(1),\n    source: z.enum([\"extraction\", \"consolidation\", \"replay\", \"manual\", \"unknown\"]),\n  })\n  .strict();\n\nexport function parseMemoryActionType(value: unknown): z.infer<typeof MemoryActionTypeSchema> {\n  const parsed = MemoryActionTypeSchema.safeParse(value);\n  return parsed.success ? parsed.data : \"discard\";\n}\n\nexport function parseMemoryActionEligibilityContext(\n  value: unknown,\n): z.infer<typeof MemoryActionEligibilityContextSchema> {\n  const parsed = MemoryActionEligibilityContextSchema.safeParse(value);\n  if (parsed.success) return parsed.data;\n  return {\n    confidence: 0,\n    lifecycleState: \"candidate\",\n    importance: 0,\n    source: \"unknown\",\n  };\n}\n\nexport const ProcedureStepExtractSchema = z.object({\n  order: z.number(),\n  intent: z.string(),\n  toolCall: z\n    .object({\n      kind: z.string(),\n      signature: z.string(),\n    })\n    .optional()\n    .nullable(),\n  expectedOutcome: z.string().optional().nullable(),\n  optional: z.boolean().optional().nullable(),\n});\n\nexport const ExtractedFactSchema = z.object({\n  category: z.enum([\n    \"fact\",\n    \"preference\",\n    \"correction\",\n    \"entity\",\n    \"decision\",\n    \"relationship\",\n    \"principle\",\n    \"commitment\",\n    \"moment\",\n    \"skill\",\n    \"rule\",\n    \"procedure\",\n    \"reasoning_trace\",\n  ]),\n  content: z\n    .string()\n    .describe(\"The memory content — a clear, standalone statement\"),\n  confidence: z\n    .number()\n    .min(0)\n    .max(1)\n    .describe(\"How confident are you this is correct (0-1)\"),\n  tags: z.array(z.string()).describe(\"Relevant tags for categorization\"),\n  entityRef: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"If about an entity, its normalized name (e.g. person-jane-doe)\"),\n  promptedByQuestion: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"Optional proactive follow-up question that surfaced this fact.\"),\n  scope: z\n    .enum([\"project\", \"global\"])\n    .optional()\n    .nullable()\n    .describe(\n      'Scope classification: \"global\" for cross-project knowledge (framework bugs, library behavior, API patterns, user preferences, tool configs, general coding patterns); \"project\" for project-specific knowledge (file paths, env configs, deployment details, project workarounds). Defaults to \"project\" when a coding context is active.',\n    ),\n  quote: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\n      \"The EXACT verbatim words from the conversation that support this fact. Copy a contiguous span from a single speaker turn (not a paraphrase). Cap at ~300 characters. This is the grounding evidence for downstream faithfulness verification (issue #1575).\",\n    ),\n  structuredAttributes: z\n    .record(z.string(), z.string())\n    .optional()\n    .nullable()\n    .describe(\"Structured key-value attributes when the fact contains measurable or categorical data (e.g., {\\\"price\\\": \\\"29.99\\\", \\\"color\\\": \\\"blue\\\", \\\"date\\\": \\\"2024-03-15\\\"}).\"),\n  procedureSteps: z\n    .array(ProcedureStepExtractSchema)\n    .optional()\n    .nullable()\n    .describe(\n      'For category \"procedure\" only: ordered steps (intent per step). At least two steps; include explicit trigger phrasing in content (e.g. \"When you deploy…\").',\n    ),\n  reasoningTrace: z\n    .object({\n      steps: z\n        .array(\n          z.object({\n            order: z.number(),\n            description: z.string(),\n          }),\n        )\n        // Prompts and normalizer require >=2 ordered steps; enforce it at\n        // the schema layer so gateway-path parsing rejects malformed traces\n        // rather than persisting them (local/direct-client normalization\n        // already enforces this, keeping the two paths symmetric).\n        .min(2)\n        .describe(\"Ordered reasoning steps the user walked through (require >=2).\"),\n      // Accept snake_case aliases so a loose gateway model using\n      // `final_answer` / `observed_outcome` does not fail schema parsing\n      // and drop the entire extraction result. Local/direct-client\n      // normalization already tolerates these keys; the gateway path must\n      // too. Zod's `union` keeps the parse successful either way.\n      finalAnswer: z\n        .string()\n        .optional()\n        .nullable()\n        .describe(\"The conclusion, decision, or answer the chain arrived at.\"),\n      final_answer: z\n        .string()\n        .optional()\n        .nullable()\n        .describe(\"Alias for finalAnswer (snake_case). Gateway-tolerance shim.\"),\n      observedOutcome: z\n        .string()\n        .optional()\n        .nullable()\n        .describe(\"Optional note about how the answer actually played out.\"),\n      observed_outcome: z\n        .string()\n        .optional()\n        .nullable()\n        .describe(\"Alias for observedOutcome (snake_case). Gateway-tolerance shim.\"),\n    })\n    // Either finalAnswer OR final_answer must be present — enforce here so\n    // the rest of the pipeline can assume a usable string downstream.\n    .refine(\n      (v) =>\n        (typeof v.finalAnswer === \"string\" && v.finalAnswer.trim().length > 0) ||\n        (typeof v.final_answer === \"string\" && v.final_answer.trim().length > 0),\n      { message: \"reasoningTrace requires finalAnswer (or final_answer)\" },\n    )\n    .optional()\n    .nullable()\n    .describe(\n      'For category \"reasoning_trace\" only: a stored solution chain with ordered steps, a final answer, and an optional observed outcome. Require at least two steps.',\n    ),\n  eventTime: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\n      'Optional event-time expression for bi-temporal validity (#1578). An ISO date (\"2025-03-01\") or a relative expression verbatim (\"last March\", \"yesterday\", \"since 2024\", \"until 2025-06-01\"). Resolved against the source turn timestamp at write time — NOT wall-clock. Omit when the fact has no explicit temporal anchor.',\n    ),\n  event_time: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"Alias for eventTime (snake_case). Gateway-tolerance shim (#1578).\"),\n}).superRefine((value, ctx) => {\n  if (value.category === \"procedure\" && (!Array.isArray(value.procedureSteps) || value.procedureSteps.length < 2)) {\n    ctx.addIssue({\n      code: z.ZodIssueCode.custom,\n      path: [\"procedureSteps\"],\n      message: \"procedure facts require at least two procedureSteps\",\n    });\n  }\n\n  if (value.category === \"reasoning_trace\" && value.reasoningTrace == null) {\n    ctx.addIssue({\n      code: z.ZodIssueCode.custom,\n      path: [\"reasoningTrace\"],\n      message: \"reasoning_trace facts require reasoningTrace\",\n    });\n  }\n});\n\nexport const EntityMentionSchema = z.object({\n  name: z\n    .string()\n    .describe(\"Normalized entity name (e.g. jane-doe, acme-corp, my-project)\"),\n  type: z.enum([\"person\", \"project\", \"tool\", \"company\", \"place\", \"other\"]),\n  facts: z\n    .array(z.string())\n    .describe(\"New facts learned about this entity in this conversation\"),\n  promptedByQuestion: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"Optional proactive follow-up question that surfaced this entity.\"),\n  structuredSections: z\n    .array(z.object({\n      key: z.string(),\n      title: z.string(),\n      facts: z.array(z.string()),\n    }))\n    .optional()\n    .nullable()\n    .describe(\"Optional named sections for entity-specific facts. Use when facts clearly belong under a durable heading such as Beliefs or Building / Working On.\"),\n});\n\nexport const ExtractedQuestionSchema = z.object({\n  question: z.string().describe(\"A genuine question the AI is curious about based on this conversation\"),\n  context: z.string().describe(\"Why this question matters or what prompted it\"),\n  priority: z.number().min(0).max(1).describe(\"How important/urgent this question is (0-1)\"),\n});\n\nexport const ProactiveQuestionsResultSchema = z.object({\n  questions: z\n    .array(ExtractedQuestionSchema)\n    .describe(\"Additional follow-up questions discovered in a proactive second-pass extraction.\"),\n});\n\nexport const ExtractedRelationshipSchema = z.object({\n  source: z.string().describe(\"Source entity name (normalized, e.g. person-jane-doe)\"),\n  target: z.string().describe(\"Target entity name (normalized, e.g. company-acme-corp)\"),\n  label: z.string().describe(\"Relationship label (e.g. 'works at', 'created', 'manages')\"),\n  promptedByQuestion: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"Optional proactive follow-up question that surfaced this relationship.\"),\n});\n\nexport const ProactiveExtractionResultSchema = z.object({\n  facts: z\n    .array(ExtractedFactSchema)\n    .describe(\n      \"Additional high-confidence memories recovered only after answering proactive follow-up questions from the same buffered conversation.\",\n    ),\n  profileUpdates: z\n    .array(z.string())\n    .describe(\n      \"Additional profile updates directly supported by the buffered conversation. Omit anything speculative.\",\n    ),\n  entities: z\n    .array(EntityMentionSchema)\n    .describe(\n      \"Additional entities or entity facts surfaced by the proactive follow-up pass.\",\n    ),\n  relationships: z\n    .array(ExtractedRelationshipSchema)\n    .optional()\n    .nullable()\n    .describe(\n      \"Additional relationships surfaced by the proactive follow-up pass.\",\n    ),\n});\n\nexport const ExtractionResultSchema = z.object({\n  facts: z\n    .array(ExtractedFactSchema)\n    .describe(\n      \"Extracted memories from the conversation. Include facts, preferences, corrections, and decisions. Only extract genuinely new, durable information — skip transient task state.\",\n    ),\n  profileUpdates: z\n    .array(z.string())\n    .describe(\n      \"Updates to the user's behavioral profile. Each string is a standalone statement about the user's preferences, habits, or personality. Only include genuinely new insights.\",\n    ),\n  entities: z\n    .array(EntityMentionSchema)\n    .describe(\n      \"Entities mentioned in the conversation with new facts about them.\",\n    ),\n  questions: z\n    .array(ExtractedQuestionSchema)\n    .describe(\n      \"1-3 genuine questions you're curious about from this conversation. These should be things you'd actually want to know the answer to in future sessions.\",\n    ),\n  identityReflection: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\n      \"A brief reflection on what you learned about yourself as an agent in this interaction — patterns in your behavior, growth, things you did well or could improve.\",\n    ),\n  relationships: z\n    .array(ExtractedRelationshipSchema)\n    .optional()\n    .nullable()\n    .describe(\n      \"Relationships between entities discovered in this conversation. Max 5 per extraction. Format: {source, target, label}.\",\n    ),\n});\n\nexport const ConsolidationItemSchema = z.object({\n  existingId: z\n    .string()\n    .describe(\"The ID of the existing memory being evaluated\"),\n  action: z.enum([\"ADD\", \"MERGE\", \"UPDATE\", \"INVALIDATE\", \"SKIP\"]),\n  mergeWith: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"If MERGE, the ID of the memory to merge with\"),\n  updatedContent: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"If UPDATE or MERGE, the new content\"),\n  reason: z.string().describe(\"Brief reason for this decision\"),\n});\n\nexport const ConsolidationResultSchema = z.object({\n  items: z\n    .array(ConsolidationItemSchema)\n    .describe(\n      \"Decisions for each existing memory: ADD (keep as-is), MERGE (combine with another), UPDATE (revise content), INVALIDATE (mark as outdated/wrong), SKIP (no action needed)\",\n    ),\n  profileUpdates: z\n    .array(z.string())\n    .describe(\"New profile statements to add or update\"),\n  entityUpdates: z\n    .array(EntityMentionSchema)\n    .describe(\"Entity updates from consolidation analysis\"),\n});\n\nexport function buildProfileConsolidationResultSchema(targetLines: number) {\n  return z.object({\n    consolidatedProfile: z\n      .string()\n      .describe(\n        `The full consolidated profile as markdown. Preserve all ## section headers. Merge duplicate or near-duplicate bullets into single clear statements. Remove stale or superseded information. Keep the most important and durable observations. Target roughly ${targetLines} lines.`,\n      ),\n    removedCount: z\n      .number()\n      .describe(\"Number of bullets removed or merged during consolidation\"),\n    summary: z\n      .string()\n      .describe(\"Brief summary of what was consolidated\"),\n  });\n}\n\nexport const ProfileConsolidationResultSchema = buildProfileConsolidationResultSchema(50);\n\nexport const IdentityConsolidationResultSchema = z.object({\n  learnedPatterns: z\n    .array(z.string())\n    .describe(\n      \"Consolidated behavioral patterns and lessons learned, each a concise standalone statement\",\n    ),\n  summary: z\n    .string()\n    .describe(\n      \"A brief paragraph summarizing the agent's core identity insights\",\n    ),\n});\n\nexport type IdentityConsolidationResultParsed = z.infer<\n  typeof IdentityConsolidationResultSchema\n>;\n\n// Contradiction Verification (Phase 2B)\nexport const ContradictionVerificationSchema = z.object({\n  isContradiction: z\n    .boolean()\n    .describe(\"Whether the two memories truly contradict each other\"),\n  confidence: z\n    .number()\n    .min(0)\n    .max(1)\n    .describe(\"How confident are you in this assessment (0-1)\"),\n  reasoning: z\n    .string()\n    .describe(\"Explanation of why these are or are not contradictory\"),\n  whichIsNewer: z\n    .enum([\"first\", \"second\", \"unclear\"])\n    .describe(\"Which memory represents the more recent/current state\"),\n});\n\nexport type ContradictionVerificationResult = z.infer<\n  typeof ContradictionVerificationSchema\n>;\n\n// Memory Linking (Phase 3A)\nexport const MemoryLinkSchema = z.object({\n  targetId: z\n    .string()\n    .describe(\"The ID of the memory this links to\"),\n  linkType: z\n    .enum([\"follows\", \"references\", \"contradicts\", \"supports\", \"related\"])\n    .describe(\"The type of relationship\"),\n  strength: z\n    .number()\n    .min(0)\n    .max(1)\n    .describe(\"How strong is this relationship (0-1)\"),\n  reason: z\n    .string()\n    .optional()\n    .nullable()\n    .describe(\"Why this link exists\"),\n});\n\nexport const SuggestedLinksSchema = z.object({\n  links: z\n    .array(MemoryLinkSchema)\n    .describe(\"Suggested links between memories based on semantic analysis\"),\n});\n\nexport type MemoryLink = z.infer<typeof MemoryLinkSchema>;\nexport type SuggestedLinks = z.infer<typeof SuggestedLinksSchema>;\n\n// Memory Summarization (Phase 4A)\nexport const MemorySummarySchema = z.object({\n  summaryText: z\n    .string()\n    .describe(\"A concise summary of the batch of memories\"),\n  keyFacts: z\n    .array(z.string())\n    .describe(\"The most important facts extracted from these memories\"),\n  keyEntities: z\n    .array(z.string())\n    .describe(\"Key entities mentioned across these memories\"),\n});\n\nexport type MemorySummaryResult = z.infer<typeof MemorySummarySchema>;\n\nexport const DaySummaryResultSchema = z.object({\n  summary: z.string().min(1).describe(\"A concise end-of-day summary paragraph.\"),\n  bullets: z.array(z.string()).default([]).describe(\"The most important moments from the day.\"),\n  next_actions: z.array(z.string()).default([]).describe(\"Concrete next actions for tomorrow.\"),\n  risks_or_open_loops: z.array(z.string()).default([]).describe(\"Open loops, blockers, or fragile assumptions still needing attention.\"),\n});\n\n// v8.15 behavior-loop auto-tuning state contracts\nexport const BehaviorLoopAdjustmentSchema = z.object({\n  parameter: z.string().min(1),\n  previousValue: z.number(),\n  nextValue: z.number(),\n  delta: z.number(),\n  evidenceCount: z.number().int().min(0),\n  confidence: z.number().min(0).max(1),\n  reason: z.string(),\n  appliedAt: z.string(),\n});\n\nexport const BehaviorLoopPolicyStateSchema = z.object({\n  version: z.number().int().min(0),\n  windowDays: z.number().int().min(0),\n  minSignalCount: z.number().int().min(0),\n  maxDeltaPerCycle: z.number().min(0).max(1),\n  protectedParams: z.array(z.string()),\n  adjustments: z.array(BehaviorLoopAdjustmentSchema),\n  updatedAt: z.string(),\n});\n\nexport type BehaviorLoopAdjustmentParsed = z.infer<typeof BehaviorLoopAdjustmentSchema>;\nexport type BehaviorLoopPolicyStateParsed = z.infer<typeof BehaviorLoopPolicyStateSchema>;\n\nexport type MemoryActionTypeParsed = z.infer<typeof MemoryActionTypeSchema>;\nexport type MemoryActionEligibilityContextParsed = z.infer<typeof MemoryActionEligibilityContextSchema>;\nexport type ExtractedFactParsed = z.infer<typeof ExtractedFactSchema>;\nexport type EntityMentionParsed = z.infer<typeof EntityMentionSchema>;\nexport type ExtractedQuestionParsed = z.infer<typeof ExtractedQuestionSchema>;\nexport type ProactiveQuestionsResultParsed = z.infer<typeof ProactiveQuestionsResultSchema>;\nexport type ExtractionResultParsed = z.infer<typeof ExtractionResultSchema>;\nexport type ConsolidationItemParsed = z.infer<typeof ConsolidationItemSchema>;\nexport type ConsolidationResultParsed = z.infer<\n  typeof ConsolidationResultSchema\n>;\n"],"mappings":";AAAA,SAAS,SAAS;AAEX,IAAM,yBAAyB,EAAE,KAAK;AAAA,EAC3C;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAEM,IAAM,uCAAuC,EACjD,OAAO;AAAA,EACN,YAAY,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EACnC,gBAAgB,EAAE,KAAK,CAAC,UAAU,aAAa,aAAa,SAAS,UAAU,CAAC;AAAA,EAChF,YAAY,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EACnC,QAAQ,EAAE,KAAK,CAAC,cAAc,iBAAiB,UAAU,UAAU,SAAS,CAAC;AAC/E,CAAC,EACA,OAAO;AAEH,SAAS,sBAAsB,OAAwD;AAC5F,QAAM,SAAS,uBAAuB,UAAU,KAAK;AACrD,SAAO,OAAO,UAAU,OAAO,OAAO;AACxC;AAEO,SAAS,oCACd,OACsD;AACtD,QAAM,SAAS,qCAAqC,UAAU,KAAK;AACnE,MAAI,OAAO,QAAS,QAAO,OAAO;AAClC,SAAO;AAAA,IACL,YAAY;AAAA,IACZ,gBAAgB;AAAA,IAChB,YAAY;AAAA,IACZ,QAAQ;AAAA,EACV;AACF;AAEO,IAAM,6BAA6B,EAAE,OAAO;AAAA,EACjD,OAAO,EAAE,OAAO;AAAA,EAChB,QAAQ,EAAE,OAAO;AAAA,EACjB,UAAU,EACP,OAAO;AAAA,IACN,MAAM,EAAE,OAAO;AAAA,IACf,WAAW,EAAE,OAAO;AAAA,EACtB,CAAC,EACA,SAAS,EACT,SAAS;AAAA,EACZ,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,SAAS;AAAA,EAChD,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS;AAC5C,CAAC;AAEM,IAAM,sBAAsB,EAAE,OAAO;AAAA,EAC1C,UAAU,EAAE,KAAK;AAAA,IACf;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAAA,EACD,SAAS,EACN,OAAO,EACP,SAAS,yDAAoD;AAAA,EAChE,YAAY,EACT,OAAO,EACP,IAAI,CAAC,EACL,IAAI,CAAC,EACL,SAAS,6CAA6C;AAAA,EACzD,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,kCAAkC;AAAA,EACrE,WAAW,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,gEAAgE;AAAA,EAC5E,oBAAoB,EACjB,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,gEAAgE;AAAA,EAC5E,OAAO,EACJ,KAAK,CAAC,WAAW,QAAQ,CAAC,EAC1B,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,OAAO,EACJ,OAAO,EACP,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,sBAAsB,EACnB,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC,EAC7B,SAAS,EACT,SAAS,EACT,SAAS,0JAAsK;AAAA,EAClL,gBAAgB,EACb,MAAM,0BAA0B,EAChC,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,gBAAgB,EACb,OAAO;AAAA,IACN,OAAO,EACJ;AAAA,MACC,EAAE,OAAO;AAAA,QACP,OAAO,EAAE,OAAO;AAAA,QAChB,aAAa,EAAE,OAAO;AAAA,MACxB,CAAC;AAAA,IACH,EAKC,IAAI,CAAC,EACL,SAAS,gEAAgE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,IAM5E,aAAa,EACV,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,2DAA2D;AAAA,IACvE,cAAc,EACX,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,6DAA6D;AAAA,IACzE,iBAAiB,EACd,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,yDAAyD;AAAA,IACrE,kBAAkB,EACf,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,iEAAiE;AAAA,EAC/E,CAAC,EAGA;AAAA,IACC,CAAC,MACE,OAAO,EAAE,gBAAgB,YAAY,EAAE,YAAY,KAAK,EAAE,SAAS,KACnE,OAAO,EAAE,iBAAiB,YAAY,EAAE,aAAa,KAAK,EAAE,SAAS;AAAA,IACxE,EAAE,SAAS,wDAAwD;AAAA,EACrE,EACC,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,WAAW,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,YAAY,EACT,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,mEAAmE;AACjF,CAAC,EAAE,YAAY,CAAC,OAAO,QAAQ;AAC7B,MAAI,MAAM,aAAa,gBAAgB,CAAC,MAAM,QAAQ,MAAM,cAAc,KAAK,MAAM,eAAe,SAAS,IAAI;AAC/G,QAAI,SAAS;AAAA,MACX,MAAM,EAAE,aAAa;AAAA,MACrB,MAAM,CAAC,gBAAgB;AAAA,MACvB,SAAS;AAAA,IACX,CAAC;AAAA,EACH;AAEA,MAAI,MAAM,aAAa,qBAAqB,MAAM,kBAAkB,MAAM;AACxE,QAAI,SAAS;AAAA,MACX,MAAM,EAAE,aAAa;AAAA,MACrB,MAAM,CAAC,gBAAgB;AAAA,MACvB,SAAS;AAAA,IACX,CAAC;AAAA,EACH;AACF,CAAC;AAEM,IAAM,sBAAsB,EAAE,OAAO;AAAA,EAC1C,MAAM,EACH,OAAO,EACP,SAAS,+DAA+D;AAAA,EAC3E,MAAM,EAAE,KAAK,CAAC,UAAU,WAAW,QAAQ,WAAW,SAAS,OAAO,CAAC;AAAA,EACvE,OAAO,EACJ,MAAM,EAAE,OAAO,CAAC,EAChB,SAAS,0DAA0D;AAAA,EACtE,oBAAoB,EACjB,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,kEAAkE;AAAA,EAC9E,oBAAoB,EACjB,MAAM,EAAE,OAAO;AAAA,IACd,KAAK,EAAE,OAAO;AAAA,IACd,OAAO,EAAE,OAAO;AAAA,IAChB,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAC3B,CAAC,CAAC,EACD,SAAS,EACT,SAAS,EACT,SAAS,oJAAoJ;AAClK,CAAC;AAEM,IAAM,0BAA0B,EAAE,OAAO;AAAA,EAC9C,UAAU,EAAE,OAAO,EAAE,SAAS,uEAAuE;AAAA,EACrG,SAAS,EAAE,OAAO,EAAE,SAAS,+CAA+C;AAAA,EAC5E,UAAU,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE,SAAS,6CAA6C;AAC3F,CAAC;AAEM,IAAM,iCAAiC,EAAE,OAAO;AAAA,EACrD,WAAW,EACR,MAAM,uBAAuB,EAC7B,SAAS,kFAAkF;AAChG,CAAC;AAEM,IAAM,8BAA8B,EAAE,OAAO;AAAA,EAClD,QAAQ,EAAE,OAAO,EAAE,SAAS,uDAAuD;AAAA,EACnF,QAAQ,EAAE,OAAO,EAAE,SAAS,yDAAyD;AAAA,EACrF,OAAO,EAAE,OAAO,EAAE,SAAS,4DAA4D;AAAA,EACvF,oBAAoB,EACjB,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,wEAAwE;AACtF,CAAC;AAEM,IAAM,kCAAkC,EAAE,OAAO;AAAA,EACtD,OAAO,EACJ,MAAM,mBAAmB,EACzB;AAAA,IACC;AAAA,EACF;AAAA,EACF,gBAAgB,EACb,MAAM,EAAE,OAAO,CAAC,EAChB;AAAA,IACC;AAAA,EACF;AAAA,EACF,UAAU,EACP,MAAM,mBAAmB,EACzB;AAAA,IACC;AAAA,EACF;AAAA,EACF,eAAe,EACZ,MAAM,2BAA2B,EACjC,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AACJ,CAAC;AAEM,IAAM,yBAAyB,EAAE,OAAO;AAAA,EAC7C,OAAO,EACJ,MAAM,mBAAmB,EACzB;AAAA,IACC;AAAA,EACF;AAAA,EACF,gBAAgB,EACb,MAAM,EAAE,OAAO,CAAC,EAChB;AAAA,IACC;AAAA,EACF;AAAA,EACF,UAAU,EACP,MAAM,mBAAmB,EACzB;AAAA,IACC;AAAA,EACF;AAAA,EACF,WAAW,EACR,MAAM,uBAAuB,EAC7B;AAAA,IACC;AAAA,EACF;AAAA,EACF,oBAAoB,EACjB,OAAO,EACP,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AAAA,EACF,eAAe,EACZ,MAAM,2BAA2B,EACjC,SAAS,EACT,SAAS,EACT;AAAA,IACC;AAAA,EACF;AACJ,CAAC;AAEM,IAAM,0BAA0B,EAAE,OAAO;AAAA,EAC9C,YAAY,EACT,OAAO,EACP,SAAS,+CAA+C;AAAA,EAC3D,QAAQ,EAAE,KAAK,CAAC,OAAO,SAAS,UAAU,cAAc,MAAM,CAAC;AAAA,EAC/D,WAAW,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,8CAA8C;AAAA,EAC1D,gBAAgB,EACb,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,qCAAqC;AAAA,EACjD,QAAQ,EAAE,OAAO,EAAE,SAAS,gCAAgC;AAC9D,CAAC;AAEM,IAAM,4BAA4B,EAAE,OAAO;AAAA,EAChD,OAAO,EACJ,MAAM,uBAAuB,EAC7B;AAAA,IACC;AAAA,EACF;AAAA,EACF,gBAAgB,EACb,MAAM,EAAE,OAAO,CAAC,EAChB,SAAS,yCAAyC;AAAA,EACrD,eAAe,EACZ,MAAM,mBAAmB,EACzB,SAAS,4CAA4C;AAC1D,CAAC;AAEM,SAAS,sCAAsC,aAAqB;AACzE,SAAO,EAAE,OAAO;AAAA,IACd,qBAAqB,EAClB,OAAO,EACP;AAAA,MACC,gQAAgQ,WAAW;AAAA,IAC7Q;AAAA,IACF,cAAc,EACX,OAAO,EACP,SAAS,0DAA0D;AAAA,IACtE,SAAS,EACN,OAAO,EACP,SAAS,wCAAwC;AAAA,EACtD,CAAC;AACH;AAEO,IAAM,mCAAmC,sCAAsC,EAAE;AAEjF,IAAM,oCAAoC,EAAE,OAAO;AAAA,EACxD,iBAAiB,EACd,MAAM,EAAE,OAAO,CAAC,EAChB;AAAA,IACC;AAAA,EACF;AAAA,EACF,SAAS,EACN,OAAO,EACP;AAAA,IACC;AAAA,EACF;AACJ,CAAC;AAOM,IAAM,kCAAkC,EAAE,OAAO;AAAA,EACtD,iBAAiB,EACd,QAAQ,EACR,SAAS,sDAAsD;AAAA,EAClE,YAAY,EACT,OAAO,EACP,IAAI,CAAC,EACL,IAAI,CAAC,EACL,SAAS,gDAAgD;AAAA,EAC5D,WAAW,EACR,OAAO,EACP,SAAS,uDAAuD;AAAA,EACnE,cAAc,EACX,KAAK,CAAC,SAAS,UAAU,SAAS,CAAC,EACnC,SAAS,uDAAuD;AACrE,CAAC;AAOM,IAAM,mBAAmB,EAAE,OAAO;AAAA,EACvC,UAAU,EACP,OAAO,EACP,SAAS,oCAAoC;AAAA,EAChD,UAAU,EACP,KAAK,CAAC,WAAW,cAAc,eAAe,YAAY,SAAS,CAAC,EACpE,SAAS,0BAA0B;AAAA,EACtC,UAAU,EACP,OAAO,EACP,IAAI,CAAC,EACL,IAAI,CAAC,EACL,SAAS,uCAAuC;AAAA,EACnD,QAAQ,EACL,OAAO,EACP,SAAS,EACT,SAAS,EACT,SAAS,sBAAsB;AACpC,CAAC;AAEM,IAAM,uBAAuB,EAAE,OAAO;AAAA,EAC3C,OAAO,EACJ,MAAM,gBAAgB,EACtB,SAAS,6DAA6D;AAC3E,CAAC;AAMM,IAAM,sBAAsB,EAAE,OAAO;AAAA,EAC1C,aAAa,EACV,OAAO,EACP,SAAS,4CAA4C;AAAA,EACxD,UAAU,EACP,MAAM,EAAE,OAAO,CAAC,EAChB,SAAS,wDAAwD;AAAA,EACpE,aAAa,EACV,MAAM,EAAE,OAAO,CAAC,EAChB,SAAS,8CAA8C;AAC5D,CAAC;AAIM,IAAM,yBAAyB,EAAE,OAAO;AAAA,EAC7C,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,SAAS,yCAAyC;AAAA,EAC7E,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,EAAE,SAAS,0CAA0C;AAAA,EAC5F,cAAc,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,EAAE,SAAS,qCAAqC;AAAA,EAC5F,qBAAqB,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,EAAE,SAAS,uEAAuE;AACvI,CAAC;AAGM,IAAM,+BAA+B,EAAE,OAAO;AAAA,EACnD,WAAW,EAAE,OAAO,EAAE,IAAI,CAAC;AAAA,EAC3B,eAAe,EAAE,OAAO;AAAA,EACxB,WAAW,EAAE,OAAO;AAAA,EACpB,OAAO,EAAE,OAAO;AAAA,EAChB,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC;AAAA,EACrC,YAAY,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EACnC,QAAQ,EAAE,OAAO;AAAA,EACjB,WAAW,EAAE,OAAO;AACtB,CAAC;AAEM,IAAM,gCAAgC,EAAE,OAAO;AAAA,EACpD,SAAS,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC;AAAA,EAC/B,YAAY,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC;AAAA,EAClC,gBAAgB,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC;AAAA,EACtC,kBAAkB,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EACzC,iBAAiB,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EACnC,aAAa,EAAE,MAAM,4BAA4B;AAAA,EACjD,WAAW,EAAE,OAAO;AACtB,CAAC;","names":[]}