@happyvertical/smrt-agents 0.42.6 → 0.42.7

package/dist/smrt-knowledge.json

@@ -3,13 +3,13 @@
   "sensitiveFieldsExcluded": true,
   "generatedAt": "1970-01-01T00:00:00.000Z",
   "packageName": "@happyvertical/smrt-agents",
-  "packageVersion": "0.42.6",
+  "packageVersion": "0.42.7",
   "sourceManifestPath": "dist/manifest.json",
   "agentDocPath": "AGENTS.md",
   "sourceHashes": {
-    "manifest": "0e6d9c0a7258c9c24a051a0e1ad2cda055d8e8d9d71345a75a72f12869c9a180",
-    "packageJson": "dff1d223377fe787d07b5d163394ed1c84792f951be97ef79600d38cc3e4a4f5",
-    "agents": "737f345126de4f98e96586ac78c8b366f838296a7e85d2fa90fad4f3ed6c7ed6"
+    "manifest": "38ae62699c3b833cf82710c898aa9c41669a42d5710db5d0b6515f64ecc0cbae",
+    "packageJson": "dff9028643850a52b56741b6fa6e6128bebc8a7290a9bb5b635f43cfaa9d8aa1",
+    "agents": "b3c0257e738c0e35a218c6baf4df5959efad10dc8021b584d620b526c9ab2753"
   },
   "exports": [
     ".",
@@ -1567,5 +1567,5 @@
     "polymorphicAssociations": 0,
     "uuidColumns": 11
   },
-  "agentDoc": "# @happyvertical/smrt-agents\n\nAgent framework for autonomous actors with inter-agent messaging, interest-based object discovery, scheduling, and multi-tenant bindings.\n\n## Agent Lifecycle\n\n`initialize()` → `validate()` → `run()` → `shutdown()`\n\n- Extend `Agent` (which extends `SmrtObject`) and implement `run()`\n- Status tracking: `idle → initializing → running → error/shutdown`\n- `execute()` runs the full lifecycle automatically\n- Process signal handling is opt-in via `manageProcessSignals: true` and is intended for single-agent processes\n\n## DispatchBus — Inter-Agent Communication\n\nAgents communicate via persistent async messaging through core's DispatchBus:\n\n```typescript\n// Emitting (in any agent)\nconst bus = await this.getDispatch();\nawait bus.emit('campaign.completed', { campaignId: '123' }, { source: 'Suasor' });\n\n// Subscribing (in receiving agent)\nasync handleDispatch(payload: unknown, metadata: DispatchMetadata): Promise<void> {\n  if (metadata.type === 'campaign.completed') await this.recordRevenue(payload);\n}\nasync run() { await this.processDispatches(); } // processes via handleDispatch()\n```\n\nCLI: `smrt dispatch:list`, `dispatch:process --subscriber Fiscus`, `dispatch:retry`, `dispatch:cleanup`\n\n## Interests — Object Discovery\n\nAgents query objects they care about via declarative filters:\n\n```typescript\nconstructor(options) {\n  super({ ...options, interests: {\n    objects: { Meeting: { filter: { status: 'upcoming' }, handler: async (m) => ({ action: 'recap' }) } },\n    qualify: async (items) => items.filter(/* AI-based post-filter */),\n  }});\n}\nasync run() { for (const { type, data } of await this.interesting()) { ... } }\n```\n\n## Configuration\n\n- **File-based**: `getModuleConfig('agent-name', defaults)` from `smrt.config.ts`\n- **DB-persisted**: `saveSlotConfig(slotId, data)` for UI overrides. Persona-backed\n  agents use `AgentOptions.personaId` as the durable config owner; legacy agents\n  use the saved Agent row id.\n- **Merged**: `getMergedConfig('slotId')` — DB overrides file config\n- **UI slots**: `static uiSlots` declares admin panels (id, label, icon, order,\n  `scope`, optional versioned `settingsSchema`). Without a registered custom\n  component, `AgentSettingsForm` renders that schema.\n\n## TenantAgent — Multi-Tenant Bindings\n\nJunction table (`tenant_agents`) binding agents to tenants with permission overrides and hierarchy resolution:\n- Explicit binding: row exists for tenant (source: 'explicit')\n- Inherited: walks up tenant hierarchy (source: 'inherited')\n- Permissions: manifest defaults merged with per-tenant overrides\n\n## AgentSchedule\n\nCron-based scheduling stored in `_smrt_agent_schedules`. Fields: `agentType`, `cron`, `method` (default: 'run'), `maxConcurrent`, `timeout`. Executed by ScheduleRunner from smrt-jobs.\n\n## Lazy agent_config Resolution (issue #1161)\n\nPersisted `agent_config` snapshots env-derived values at sync time, so rotated env vars don't reach already-stored schedule rows. Two complementary mechanisms unfreeze them:\n\n1. **`$env` sentinels in persisted config** — register a global resolver and reference it from the JSON:\n\n   ```ts\n   import { registerConfigResolver } from '@happyvertical/smrt-agents';\n   registerConfigResolver('sharedAssetStorage', () => resolveSharedAssetStorage());\n   // persisted: { \"assetStorage\": { \"$env\": \"sharedAssetStorage\" } }\n   ```\n\n2. **`static configResolvers` on the agent class** — declarative, discoverable via the class itself:\n\n   ```ts\n   class Praeco extends Agent {\n     static override configResolvers = {\n       assetStorage: () => resolveSharedAssetStorage(),\n     };\n   }\n   ```\n\nThe TaskRunner calls `resolveLazyConfig()` immediately before constructing the agent, so live values always win over snapshotted ones. Re-exported from `@happyvertical/smrt-core` (`resolveLazyConfig`, `registerConfigResolver`, `getClassConfigResolvers`, …) for cases where agents isn't on the import path.\n\n## Learning Trait (issue #1886) — opt-in\n\nAny agent can opt into a confidence-scored **recall-before / capture-after** loop backed by core's `LearningMemory` (over `_smrt_contexts` + `_smrt_embeddings`). **Off by default** — a non-opted agent behaves byte-for-byte as today; the lifecycle's learning branches are never entered.\n\n```typescript\n@smrt()\nclass InvoiceAgent extends Agent {\n  static override learning = true; // or { minConfidence: 0.8, scope: 'invoices', ... }\n  protected config = {};\n\n  async run() {\n    // recall-before-run already populated `recalledMemories` (confidence >= floor)\n    const cached = this.recalledMemories.find((m) => m.key === this.docUrl);\n    const strategy = cached?.value ?? (await this.generateStrategy());\n\n    // stage the episode; the lifecycle reinforces it after run()\n    this.stageLearning({ scope: this.learningScope(), key: this.docUrl, value: strategy });\n\n    // a validated failure decays the memory without throwing\n    if (!ok) this.reportLearningOutcome({ success: false, error: 'no match' });\n  }\n}\n```\n\n- **`capture` semantics** (`LearningMemory`): success strengthens `confidence` toward 1.0 and increments `success_count`; failure decays toward `failureConfidence` (0.3) and increments `failure_count`. A single failure drops a confident memory below the reuse floor (0.7), so recall stops returning it. Refreshes `last_used_at`; honours `expires_at` and optional time-decay.\n- **Memory isolation**: bound to `(agentType, agentInstanceId)` as `(owner_class, owner_id)`, so two tenants on the same agent class never share memory. `tenantId` is threaded into the optional semantic-search `where`.\n- **Seams to override**: `learningScope()`, `recallForRun(memory)`, `captureForRun(memory, outcome)`, `getLearningSemanticSearch()`. Helpers for `run()`: `stageLearning(episode)`, `reportLearningOutcome(outcome)`, `getLearningMemory()`, and the `recalledMemories` field.\n- **Config**: `static learning: boolean | AgentLearningConfig` — `{ enabled?, scope?, minConfidence?, successConfidence?, failureConfidence?, reinforcement?, decayHalfLifeMs? }`. `LearningMemory` and its types are re-exported from `@happyvertical/smrt-core`.\n\n## Multi-Instance Agents (issue #1890) — opt-in\n\n`static multiInstance = false` by default: a class is a **singleton** (the N=1 case) and is byte-for-byte unchanged — one dispatch subscriber keyed by the agent type, one memory scope, class-wide interests. Set `static multiInstance = true` to run N durable instances (personas, from `@happyvertical/smrt-personas`) of one class per tenant, each independent.\n\nThe framework provides only the per-instance **identity**; a package scopes its own dispatch/interests to the instance's config by overriding the seams.\n\n- **`AgentOptions.personaId`** — durable persona/settings identity, including\n  the default persona. Independent from execution instance identity.\n- **`AgentOptions.instanceKey`** — the durable execution-instance key (typically\n  the persona id, but null for the default persona). Honored **only** when\n  `multiInstance` is true, so passing it to a non-opted agent is a no-op.\n- **`getInstanceKey()`** → the key, or `null` for a singleton (opt-in off, or no key).\n- **`getDispatchSubscriber()`** → `` `${agentType}#${key}` `` for a multi-instance agent, the bare `agentType` for a singleton. Used everywhere the agent subscribes/seeds/processes, so each instance has its own subscription rows and pending-dispatch queue — two instances never compete for or double-process each other's dispatches. Composed by the exported `instanceScopedSubscriber(agentType, key)`.\n- **`learningScope()`** — suffixed with `#<key>` for a multi-instance agent, so instances learn independently (singleton scope unchanged).\n- **Seams to override** (both default to singleton behavior):\n  - `resolveSignalSubscriptions()` — derive **instance-scoped** signal types from the instance config so an emit meant for one instance only matches its subscription.\n  - `instanceInterestFilter()` — an `ObjectFilter` AND-merged (as the base layer) into every `interesting()` query so instances partition the objects they process.\n\nThe **`default` persona reuses the singleton identity** (a `null` key), which is what makes the singleton→multi upgrade non-destructive — see `@happyvertical/smrt-personas` (`personaInstanceKey`, `upgradeSingletonToDefaultPersona`).\n\n## Principal Execution (issue #1888)\n\n`executeAsPrincipal(options, fn)` runs agent work **AS a persona's bound user**, reusing the existing RBAC cascade with no snapshotting. It publishes `(user_id, tenant_id, permissions[])` onto the DB session (Postgres RLS then bounds every query per-`(table, action)` and per-tenant) and hands `fn` a `PrincipalRun` whose `assertToolAllowed()` / `assertOperation()` enforce the persona tool ceiling and the RLS-off catalog gate. Effective authority = **bound-user RBAC ∩ agent-class ceiling ∩ persona `allowedTools`**. Actions audit as on-behalf-of the originating user via a `PrincipalAuditSink`.\n\n## Agent Orchestration (issue #1892) — invoke-agent + principal delegation\n\nA conversational (orchestrator) agent can invoke worker agents with **principal delegation**. This is *not* a new engine — it is a standard `invoke-agent` tool plus a completion-dispatch convention on top of `executeAsPrincipal` + the DispatchBus.\n\n```typescript\nimport { createInvokeAgentTool, rootDelegationEnvelope } from '@happyvertical/smrt-agents';\n\nconst tool = createInvokeAgentTool({\n  db,\n  parentEnvelope: rootDelegationEnvelope({ runAsUserId, tenantId, onBehalfOfUserId }),\n  worker: async ({ run, agentClass, task }) => runWorker(run, agentClass, task),\n});\n// Offered through the chat tool loop as an `extraTools` entry, gated by the\n// persona's allowedTools like any other tool (slug: 'agents.invoke').\n```\n\n- **`DelegationEnvelope`** carries the **immutable principal** (`runAsUserId` + `tenantId` + originating `onBehalfOfUserId`) and a bounded `depth`. `deriveDelegationEnvelope()` copies the principal verbatim and asserts `depth <= MAX_DELEGATION_DEPTH` (3) — a worker cannot invoke a further worker under a broader principal (`PrincipalWideningError` / `DelegationDepthExceededError`).\n- **`createInvokeAgentTool()`** → a `PrincipalTool` whose handler derives the child envelope with the principal taken **from the live run context, never the tool args**, so the invoke-agent tool is structurally immune to principal widening.\n- **`executeDelegatedInvocation()`** runs the worker via `executeAsPrincipal` under that same principal and emits a correlated `agent.completed` dispatch; **`surfaceAgentCompletions(bus, correlationId)`** reads it back into the conversation.\n- **Transports** (pluggable): the default `inlineInvokeAgentTransport` runs the worker in-process (completion surfaces in the same turn); `createDispatchInvokeTransport(bus)` emits an `agent.invoke` signal a worker processes via `processAgentInvocations()` (async). A job-queue transport (enqueue on the `agents` queue) is a consumer-supplied `InvokeAgentTransport` — orchestration never hard-depends on `@happyvertical/smrt-jobs`, which sits *below* agents in the dependency graph.\n\n## Key Files\n\n| File | Purpose |\n|------|---------|\n| `src/agent.ts` | Base Agent class — lifecycle, dispatch, interests, config, opt-in learning trait, multi-instance identity |\n| `src/execute-as-principal.ts` | `executeAsPrincipal` / `PrincipalRun` — run agent work as a persona's bound user (#1888) |\n| `src/delegation.ts` | `DelegationEnvelope` — immutable principal + bounded delegation depth (#1892) |\n| `src/invoke-agent.ts` | `invoke-agent` tool, worker executor, completion-dispatch convention, transports (#1892) |\n| `src/learning.ts` | `AgentLearningConfig` + `resolveAgentLearning()` declaration normalisation |\n| `src/schedule.ts` | AgentSchedule model — cron, execution tracking |\n| `src/tenant-agent.ts` | TenantAgent — junction table, hierarchical resolution |\n| `src/interests.ts` | Interest filter types and configuration |\n| `src/config.ts` | File + DB config management, UI slots |\n"
+  "agentDoc": "# @happyvertical/smrt-agents\n\nAgent framework for autonomous actors with inter-agent messaging, interest-based object discovery, scheduling, and multi-tenant bindings.\n\n## Agent Lifecycle\n\n`initialize()` → `validate()` → `run()` → `shutdown()`\n\n- Extend `Agent` (which extends `SmrtObject`) and implement `run()`\n- Status tracking: `idle → initializing → running → error/shutdown`\n- `execute()` runs the full lifecycle automatically\n- Process signal handling is opt-in via `manageProcessSignals: true` and is intended for single-agent processes\n\n## DispatchBus — Inter-Agent Communication\n\nAgents communicate via persistent async messaging through core's DispatchBus:\n\n```typescript\n// Emitting (in any agent)\nconst bus = await this.getDispatch();\nawait bus.emit('campaign.completed', { campaignId: '123' }, { source: 'Suasor' });\n\n// Subscribing (in receiving agent)\nasync handleDispatch(payload: unknown, metadata: DispatchMetadata): Promise<void> {\n  if (metadata.type === 'campaign.completed') await this.recordRevenue(payload);\n}\nasync run() { await this.processDispatches(); } // processes via handleDispatch()\n```\n\nCLI: `smrt dispatch:list`, `dispatch:process --subscriber Fiscus`, `dispatch:retry`, `dispatch:cleanup`\n\n## Interests — Object Discovery\n\nAgents query objects they care about via declarative filters:\n\n```typescript\nconstructor(options) {\n  super({ ...options, interests: {\n    objects: { Meeting: { filter: { status: 'upcoming' }, handler: async (m) => ({ action: 'recap' }) } },\n    qualify: async (items) => items.filter(/* AI-based post-filter */),\n  }});\n}\nasync run() { for (const { type, data } of await this.interesting()) { ... } }\n```\n\n## Configuration\n\n- **File-based**: `getModuleConfig('agent-name', defaults)` from `smrt.config.ts`\n- **DB-persisted**: `saveSlotConfig(slotId, data)` for UI overrides. Persona-backed\n  agents use `AgentOptions.personaId` as the durable config owner; legacy agents\n  use the saved Agent row id.\n- **Merged**: `getMergedConfig('slotId')` — DB overrides file config\n- **UI slots**: `static uiSlots` declares admin panels (id, label, icon, order,\n  `scope`, optional versioned `settingsSchema`). Without a registered custom\n  component, `AgentSettingsForm` renders that schema.\n\n## TenantAgent — Multi-Tenant Bindings\n\nJunction table (`tenant_agents`) binding agents to tenants with permission overrides and hierarchy resolution:\n- Explicit binding: row exists for tenant (source: 'explicit')\n- Inherited: walks up tenant hierarchy (source: 'inherited')\n- Permissions: manifest defaults merged with per-tenant overrides\n\n## AgentSchedule\n\nCron-based scheduling stored in `_smrt_agent_schedules`. Fields: `agentType`, `cron`, `method` (default: 'run'), `maxConcurrent`, `timeout`. Executed by ScheduleRunner from smrt-jobs.\n\n## Lazy agent_config Resolution (issue #1161)\n\nPersisted `agent_config` snapshots env-derived values at sync time, so rotated env vars don't reach already-stored schedule rows. Two complementary mechanisms unfreeze them:\n\n1. **`$env` sentinels in persisted config** — register a global resolver and reference it from the JSON:\n\n   ```ts\n   import { registerConfigResolver } from '@happyvertical/smrt-agents';\n   registerConfigResolver('sharedAssetStorage', () => resolveSharedAssetStorage());\n   // persisted: { \"assetStorage\": { \"$env\": \"sharedAssetStorage\" } }\n   ```\n\n2. **`static configResolvers` on the agent class** — declarative, discoverable via the class itself:\n\n   ```ts\n   class Praeco extends Agent {\n     static override configResolvers = {\n       assetStorage: () => resolveSharedAssetStorage(),\n     };\n   }\n   ```\n\nThe TaskRunner calls `resolveLazyConfig()` immediately before constructing the agent, so live values always win over snapshotted ones. Re-exported from `@happyvertical/smrt-core` (`resolveLazyConfig`, `registerConfigResolver`, `getClassConfigResolvers`, …) for cases where agents isn't on the import path.\n\n## Learning Trait (issue #1886) — opt-in\n\nAny agent can opt into a confidence-scored **recall-before / capture-after** loop backed by core's `LearningMemory` (over `_smrt_contexts` + `_smrt_embeddings`). **Off by default** — a non-opted agent behaves byte-for-byte as today; the lifecycle's learning branches are never entered.\n\n```typescript\n@smrt()\nclass InvoiceAgent extends Agent {\n  static override learning = true; // or { minConfidence: 0.8, scope: 'invoices', ... }\n  protected config = {};\n\n  async run() {\n    // recall-before-run already populated `recalledMemories` (confidence >= floor)\n    const cached = this.recalledMemories.find((m) => m.key === this.docUrl);\n    const strategy = cached?.value ?? (await this.generateStrategy());\n\n    // stage the episode; the lifecycle reinforces it after run()\n    this.stageLearning({ scope: this.learningScope(), key: this.docUrl, value: strategy });\n\n    // a validated failure decays the memory without throwing\n    if (!ok) this.reportLearningOutcome({ success: false, error: 'no match' });\n  }\n}\n```\n\n- **`capture` semantics** (`LearningMemory`): success strengthens `confidence` toward 1.0 and increments `success_count`; failure decays toward `failureConfidence` (0.3) and increments `failure_count`. A single failure drops a confident memory below the reuse floor (0.7), so recall stops returning it. Refreshes `last_used_at`; honours `expires_at` and optional time-decay.\n- **Memory isolation**: bound to `(agentType, agentInstanceId)` as `(owner_class, owner_id)`, so two tenants on the same agent class never share memory. `tenantId` is threaded into the optional semantic-search `where`.\n- **Seams to override**: `learningScope()`, `recallForRun(memory)`, `captureForRun(memory, outcome)`, `getLearningSemanticSearch()`. Helpers for `run()`: `stageLearning(episode)`, `reportLearningOutcome(outcome)`, `getLearningMemory()`, and the `recalledMemories` field.\n- **Config**: `static learning: boolean | AgentLearningConfig` — `{ enabled?, scope?, minConfidence?, successConfidence?, failureConfidence?, reinforcement?, decayHalfLifeMs? }`. `LearningMemory` and its types are re-exported from `@happyvertical/smrt-core`.\n\n## Multi-Instance Agents (issue #1890) — opt-in\n\n`static multiInstance = false` by default: a class is a **singleton** (the N=1 case) and is byte-for-byte unchanged — one dispatch subscriber keyed by the agent type, one memory scope, class-wide interests. Set `static multiInstance = true` to run N durable instances (personas, from `@happyvertical/smrt-personas`) of one class per tenant, each independent.\n\nThe framework provides only the per-instance **identity**; a package scopes its own dispatch/interests to the instance's config by overriding the seams.\n\n- **`AgentOptions.personaId`** — durable persona/settings identity, including\n  the default persona. Independent from execution instance identity.\n- **`AgentOptions.instanceKey`** — the durable execution-instance key (typically\n  the persona id, but null for the default persona). Honored **only** when\n  `multiInstance` is true, so passing it to a non-opted agent is a no-op.\n- **`getInstanceKey()`** → the key, or `null` for a singleton (opt-in off, or no key).\n- **`getDispatchSubscriber()`** → `` `${agentType}#${key}` `` for a multi-instance agent, the bare `agentType` for a singleton. Used everywhere the agent subscribes/seeds/processes, so each instance has its own subscription rows and pending-dispatch queue — two instances never compete for or double-process each other's dispatches. Composed by the exported `instanceScopedSubscriber(agentType, key)`.\n- **`learningScope()`** — suffixed with `#<key>` for a multi-instance agent, so instances learn independently (singleton scope unchanged).\n- **Seams to override** (both default to singleton behavior):\n  - `resolveSignalSubscriptions()` — derive **instance-scoped** signal types from the instance config so an emit meant for one instance only matches its subscription.\n  - `instanceInterestFilter()` — an `ObjectFilter` AND-merged (as the base layer) into every `interesting()` query so instances partition the objects they process.\n\nThe **`default` persona reuses the singleton identity** (a `null` key), which is what makes the singleton→multi upgrade non-destructive — see `@happyvertical/smrt-personas` (`personaInstanceKey`, `upgradeSingletonToDefaultPersona`).\n\n## Principal Execution (issue #1888)\n\n`executeAsPrincipal(options, fn)` runs agent work **AS a persona's bound user**, reusing the existing RBAC cascade with no snapshotting. It publishes `(user_id, tenant_id, permissions[])` onto the DB session (Postgres RLS then bounds every query per-`(table, action)` and per-tenant) and hands `fn` a `PrincipalRun` whose `assertToolAllowed()` / `assertOperation()` enforce the persona tool ceiling and the RLS-off catalog gate. Effective authority = **bound-user RBAC ∩ agent-class ceiling ∩ persona `allowedTools`**. Actions audit as on-behalf-of the originating user via a `PrincipalAuditSink`.\n\n## Data Surface Read Tools (issue #2447)\n\n`createDataSurfaceTools()` produces the `data.discover`, `data.inspect`, and\n`data.query` `PrincipalTool`s consumed through chat's `extraTools` seam. The\ncaller supplies a server-owned surface catalog and executor; the tools copy\n`userId`, `tenantId`, database, and permissions only from the live\n`PrincipalRun`. Discovery and inspection first assert the collection's read\ncatalog permission and omit denied surfaces/fields. Query requests and results\nare normalized with the core bounded data-query protocol, including projection,\ncursor/page, row/byte, fingerprint, freshness, total, and truncation rules.\nSensitive/read-permission fields are removed from descriptors, and\n`DataSurfaceField` policy metadata is stripped before the core schema validator.\nExecutor-provided paginated rows retain their order and are validated with a\nstable identity tie-breaker (including type-aware numeric/date comparisons),\nwhile internal sort keys are stripped when they were not requested in the\nprojection. Execution has a bounded deadline; public executor/result failures\nuse stable generic errors while optional `onFailure` telemetry receives the\nauthenticated/delegated principal and detailed server-side error. Hidden field\nrequest failures also use a stable public error. Tool arguments never contain\nprincipal or tenant authority.\n\n## Agent Orchestration (issue #1892) — invoke-agent + principal delegation\n\nA conversational (orchestrator) agent can invoke worker agents with **principal delegation**. This is *not* a new engine — it is a standard `invoke-agent` tool plus a completion-dispatch convention on top of `executeAsPrincipal` + the DispatchBus.\n\n```typescript\nimport { createInvokeAgentTool, rootDelegationEnvelope } from '@happyvertical/smrt-agents';\n\nconst tool = createInvokeAgentTool({\n  db,\n  parentEnvelope: rootDelegationEnvelope({ runAsUserId, tenantId, onBehalfOfUserId }),\n  worker: async ({ run, agentClass, task }) => runWorker(run, agentClass, task),\n});\n// Offered through the chat tool loop as an `extraTools` entry, gated by the\n// persona's allowedTools like any other tool (slug: 'agents.invoke').\n```\n\n- **`DelegationEnvelope`** carries the **immutable principal** (`runAsUserId` + `tenantId` + originating `onBehalfOfUserId`) and a bounded `depth`. `deriveDelegationEnvelope()` copies the principal verbatim and asserts `depth <= MAX_DELEGATION_DEPTH` (3) — a worker cannot invoke a further worker under a broader principal (`PrincipalWideningError` / `DelegationDepthExceededError`).\n- **`createInvokeAgentTool()`** → a `PrincipalTool` whose handler derives the child envelope with the principal taken **from the live run context, never the tool args**, so the invoke-agent tool is structurally immune to principal widening.\n- **`executeDelegatedInvocation()`** runs the worker via `executeAsPrincipal` under that same principal and emits a correlated `agent.completed` dispatch; **`surfaceAgentCompletions(bus, correlationId)`** reads it back into the conversation.\n- **Transports** (pluggable): the default `inlineInvokeAgentTransport` runs the worker in-process (completion surfaces in the same turn); `createDispatchInvokeTransport(bus)` emits an `agent.invoke` signal a worker processes via `processAgentInvocations()` (async). A job-queue transport (enqueue on the `agents` queue) is a consumer-supplied `InvokeAgentTransport` — orchestration never hard-depends on `@happyvertical/smrt-jobs`, which sits *below* agents in the dependency graph.\n\n## Key Files\n\n| File | Purpose |\n|------|---------|\n| `src/agent.ts` | Base Agent class — lifecycle, dispatch, interests, config, opt-in learning trait, multi-instance identity |\n| `src/execute-as-principal.ts` | `executeAsPrincipal` / `PrincipalRun` — run agent work as a persona's bound user (#1888) |\n| `src/delegation.ts` | `DelegationEnvelope` — immutable principal + bounded delegation depth (#1892) |\n| `src/invoke-agent.ts` | `invoke-agent` tool, worker executor, completion-dispatch convention, transports (#1892) |\n| `src/learning.ts` | `AgentLearningConfig` + `resolveAgentLearning()` declaration normalisation |\n| `src/schedule.ts` | AgentSchedule model — cron, execution tracking |\n| `src/tenant-agent.ts` | TenantAgent — junction table, hierarchical resolution |\n| `src/interests.ts` | Interest filter types and configuration |\n| `src/config.ts` | File + DB config management, UI slots |\n"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@happyvertical/smrt-agents",
-  "version": "0.42.6",
+  "version": "0.42.7",
   "type": "module",
   "smrtRawPrimitives": "strict",
   "description": "Agent framework for building autonomous actors in the SMRT ecosystem",
@@ -59,13 +59,13 @@
     "@happyvertical/ai": "^0.88.0",
     "@happyvertical/files": "^0.88.0",
     "@happyvertical/utils": "^0.88.0",
-    "@happyvertical/smrt-config": "0.42.6",
-    "@happyvertical/smrt-core": "0.42.6",
-    "@happyvertical/smrt-secrets": "0.42.6",
-    "@happyvertical/smrt-tenancy": "0.42.6",
-    "@happyvertical/smrt-ui": "0.42.6",
-    "@happyvertical/smrt-types": "0.42.6",
-    "@happyvertical/smrt-users": "0.42.6"
+    "@happyvertical/smrt-config": "0.42.7",
+    "@happyvertical/smrt-core": "0.42.7",
+    "@happyvertical/smrt-secrets": "0.42.7",
+    "@happyvertical/smrt-ui": "0.42.7",
+    "@happyvertical/smrt-tenancy": "0.42.7",
+    "@happyvertical/smrt-types": "0.42.7",
+    "@happyvertical/smrt-users": "0.42.7"
   },
   "devDependencies": {
     "@happyvertical/logger": "^0.88.0",
@@ -82,7 +82,7 @@
     "typescript": "5.9.3",
     "vite": "8.1.4",
     "vitest": "4.1.10",
-    "@happyvertical/smrt-vitest": "0.42.6"
+    "@happyvertical/smrt-vitest": "0.42.7"
   },
   "keywords": [
     "agent",

package/dist/chunks/config-BRQLhsFp.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"config-BRQLhsFp.js","names":["tenantId"],"sources":["../../src/identity.ts","../../src/config.ts"],"sourcesContent":["import { getClassName, ObjectRegistry } from '@happyvertical/smrt-core';\n\n/**\n * Return the canonical agent type identifier for storage and dispatch routing.\n *\n * Uses the registry's qualified name when available and falls back to the input\n * name for dynamically defined or unregistered classes.\n */\nexport function getAgentTypeName(name: string): string {\n  const registered = ObjectRegistry.getClass(name);\n  return registered?.qualifiedName || registered?.name || name;\n}\n\n/**\n * Return the human-readable class name for UI and logs.\n */\nexport function getAgentClassName(name: string): string {\n  const registered = ObjectRegistry.getClass(name);\n  return registered?.name || getClassName(name);\n}\n\n/**\n * Return all meaningful aliases for an agent type.\n *\n * The qualified name is first so persistence lookups prefer canonical rows,\n * while the simple class name keeps legacy rows discoverable during migration.\n */\nexport function getAgentTypeAliases(name: string): string[] {\n  return Array.from(\n    new Set([getAgentTypeName(name), getAgentClassName(name)].filter(Boolean)),\n  );\n}\n\n/**\n * Compose a per-instance dispatch subscriber identity from an agent type and an\n * optional instance key (#1890).\n *\n * Multiple durable instances of one agent class each need their own subscriber\n * name so their dispatch subscriptions and pending dispatches never collide —\n * that is what keeps two instances from double-processing each other's work.\n *\n * Returns the bare `agentType` when `instanceKey` is nullish/empty, so a\n * **singleton** agent's subscriber is byte-for-byte unchanged (the N=1 default).\n * When a key is present the identity is `` `${agentType}#${instanceKey}` `` — a\n * stable, reversible composition (the type never contains `#`).\n */\nexport function instanceScopedSubscriber(\n  agentType: string,\n  instanceKey?: string | null,\n): string {\n  return instanceKey ? `${agentType}#${instanceKey}` : agentType;\n}\n","/**\n * AgentConfig - Persistent configuration storage for agents\n *\n * This module provides database-backed configuration for agents,\n * enabling consuming apps to persist agent settings.\n *\n * @module\n */\n\nimport {\n  field,\n  type SmrtClassOptions,\n  SmrtCollection,\n  SmrtObject,\n  smrt,\n} from '@happyvertical/smrt-core';\nimport {\n  queryGlobal,\n  queryWithGlobals,\n  TenantScoped,\n  tenantId,\n} from '@happyvertical/smrt-tenancy';\nimport { getAgentTypeName } from './identity.js';\n\n/**\n * AgentConfig stores agent configuration in the database\n *\n * Each config record maps to a UI slot for an agent configuration owner:\n * - agentId: The durable config owner ID (persona ID for persona-backed\n *   instances; Agent STI row ID for legacy/singleton instances)\n * - agentClass: The canonical agent type (qualified name when available)\n * - slotId: The configuration slot (e.g., 'sources', 'settings')\n * - configData: JSON object containing the configuration\n *\n * @example\n * ```typescript\n * // Save config for an agent slot\n * const config = new AgentConfig({\n *   agentId: agent.id,\n *   agentClass: 'Praeco',\n *   slotId: 'sources',\n *   configData: { scrapers: ['civicweb', 'govstack'] },\n *   db: options.db\n * });\n * await config.initialize();\n * await config.save();\n * ```\n */\n@TenantScoped({ mode: 'optional' })\n@smrt({\n  tableName: 'agent_configs',\n  api: { include: ['list', 'get', 'create', 'update', 'delete'] },\n  mcp: { include: ['list', 'get'] },\n  cli: true,\n})\nexport class AgentConfig extends SmrtObject {\n  /**\n   * Tenant ID for multi-tenant isolation\n   * Nullable to support both tenant-scoped and global agent configs\n   */\n  @tenantId({ nullable: true })\n  tenantId: string | null = null;\n\n  /**\n   * Durable configuration owner ID.\n   *\n   * The database column retains the historical `agent_id` name for backward\n   * compatibility. Persona-backed runtimes store their `AgentPersona.id` here;\n   * legacy runtimes store the persisted Agent STI row id.\n   */\n  @field({ type: 'text' })\n  agentId: string = '';\n\n  /**\n   * Canonical agent type for this config (qualified name when available)\n   */\n  @field({ type: 'text' })\n  agentClass: string = '';\n\n  /**\n   * UI slot ID (e.g., 'sources', 'settings', 'reports')\n   */\n  @field({ type: 'text' })\n  slotId: string = '';\n\n  /**\n   * Configuration data stored as JSON\n   *\n   * Sensitive (#1540) for backward compatibility with legacy blobs, so this is\n   * excluded from generated API/MCP responses and rejected as a `where` filter\n   * key. New settings schemas must keep credentials in a dedicated secrets\n   * service rather than this field.\n   */\n  @field({ type: 'json', sensitive: true })\n  configData: Record<string, unknown> = {};\n\n  /**\n   * Schema version for future migrations\n   */\n  @field({ type: 'integer' })\n  schemaVersion: number = 1;\n\n  /**\n   * Load all configs for a specific agent\n   *\n   * @param agentId - Agent instance ID\n   * @param options - Database options\n   * @returns Map of slotId → configData\n   */\n  static async forAgent(\n    agentId: string,\n    options: SmrtClassOptions,\n  ): Promise<Map<string, Record<string, unknown>>> {\n    const configsByAgent = await AgentConfig.forAgents([agentId], options);\n    return configsByAgent.get(agentId) ?? new Map();\n  }\n\n  /**\n   * Load configs for multiple agents in a single query.\n   *\n   * @param agentIds - Agent instance IDs\n   * @param options - Database options\n   * @returns Map of agentId -> (slotId -> configData)\n   */\n  static async forAgents(\n    agentIds: string[],\n    options: SmrtClassOptions,\n  ): Promise<Map<string, Map<string, Record<string, unknown>>>> {\n    const configsByAgent = new Map<\n      string,\n      Map<string, Record<string, unknown>>\n    >();\n    if (agentIds.length === 0) {\n      return configsByAgent;\n    }\n\n    const collection = await AgentConfigCollection.create(options);\n    const configs = await collection.list({\n      where: { 'agentId in': agentIds },\n    });\n\n    for (const config of configs) {\n      if (!configsByAgent.has(config.agentId)) {\n        configsByAgent.set(config.agentId, new Map());\n      }\n      configsByAgent.get(config.agentId)?.set(config.slotId, config.configData);\n    }\n\n    return configsByAgent;\n  }\n\n  /**\n   * Load config for a specific agent and slot\n   *\n   * @param agentId - Agent instance ID\n   * @param slotId - UI slot ID\n   * @param options - Database options\n   * @returns Config data or undefined if not found\n   */\n  static async forSlot(\n    agentId: string,\n    slotId: string,\n    options: SmrtClassOptions,\n  ): Promise<Record<string, unknown> | undefined> {\n    const collection = await AgentConfigCollection.create(options);\n    const configs = await collection.list({\n      where: { agentId, slotId },\n      limit: 1,\n    });\n    return configs[0]?.configData;\n  }\n\n  /**\n   * Save or update config for an agent slot\n   *\n   * @param data - Config data including agentId, agentClass, slotId, configData\n   * @param options - Database options\n   * @returns Saved AgentConfig instance\n   */\n  static async saveSlot(\n    data: {\n      agentId: string;\n      agentClass: string;\n      slotId: string;\n      configData: Record<string, unknown>;\n    },\n    options: SmrtClassOptions,\n  ): Promise<AgentConfig> {\n    const normalizedAgentClass = getAgentTypeName(data.agentClass);\n    const collection = await AgentConfigCollection.create(options);\n\n    // Check for existing config using list with where clause\n    const existingConfigs = await collection.list({\n      where: { agentId: data.agentId, slotId: data.slotId },\n      limit: 1,\n    });\n\n    if (existingConfigs.length > 0) {\n      // Update existing\n      const existing = existingConfigs[0];\n      existing.configData = data.configData;\n      existing.agentClass = normalizedAgentClass;\n      await existing.save();\n      return existing;\n    }\n\n    // Create new\n    const config = await collection.create({\n      agentId: data.agentId,\n      agentClass: normalizedAgentClass,\n      slotId: data.slotId,\n      configData: data.configData,\n      slug: `${data.agentId}-${data.slotId}`,\n    });\n    await config.save();\n    return config;\n  }\n}\n\n/**\n * Collection for AgentConfig objects\n */\nexport class AgentConfigCollection extends SmrtCollection<AgentConfig> {\n  static readonly _itemClass = AgentConfig;\n\n  /**\n   * Find all configs for a specific tenant\n   * @param tenantId - Tenant ID to filter by\n   * @returns Array of AgentConfig objects for the tenant\n   */\n  async findByTenant(tenantId: string): Promise<AgentConfig[]> {\n    return this.list({ where: { tenantId } });\n  }\n\n  /**\n   * Find all global configs (not associated with any tenant).\n   *\n   * Routes through the shared tenant-global helper so it does not throw under\n   * an active tenant context (an explicit `tenant_id IS NULL` filter would be\n   * flagged as an isolation violation). (#1600)\n   *\n   * @returns Array of global AgentConfig objects\n   */\n  async findGlobal(): Promise<AgentConfig[]> {\n    return queryGlobal<AgentConfig>(this);\n  }\n\n  /**\n   * Find configs for a tenant including global configs.\n   *\n   * Fails closed if an active tenant context requests a different tenant's\n   * rows; the admin/system path keeps the cross-tenant capability. (#1600)\n   *\n   * @param tenantId - Tenant ID to include\n   * @returns Array of AgentConfig objects for the tenant and global configs\n   */\n  async findWithGlobals(tenantId: string): Promise<AgentConfig[]> {\n    return queryWithGlobals<AgentConfig>(\n      this,\n      tenantId,\n      'AgentConfig.findWithGlobals',\n    );\n  }\n}\n"],"mappings":";;;AAQO,SAAS,iBAAiB,MAAsB;CACrD,MAAM,aAAa,eAAe,SAAS,IAAI;CAC/C,OAAO,YAAY,iBAAiB,YAAY,QAAQ;AAC1D;AAKO,SAAS,kBAAkB,MAAsB;CAEtD,OADmB,eAAe,SAAS,IACpC,CAAA,EAAY,QAAQ,aAAa,IAAI;AAC9C;AAQO,SAAS,oBAAoB,MAAwB;CAC1D,OAAO,MAAM,KACX,IAAI,IAAI,CAAC,iBAAiB,IAAI,GAAG,kBAAkB,IAAI,CAAC,CAAA,CAAE,OAAO,OAAO,CAAC,CAC3E;AACF;AAeO,SAAS,yBACd,WACA,aACQ;CACR,OAAO,cAAc,GAAG,UAAS,GAAI,gBAAgB;AACvD;;;;;;;;;;;ACIO,IAAM,cAAN,cAA0B,WAAW;CAM1C,WAA0B;CAU1B,UAAkB;CAMlB,aAAqB;CAMrB,SAAiB;CAWjB,aAAsC,CAAC;CAMvC,gBAAwB;;;;;;;;CASxB,aAAa,SACX,SACA,SAC+C;EAE/C,QAAO,MADsB,YAAY,UAAU,CAAC,OAAO,GAAG,OAAO,EAAA,CAC/C,IAAI,OAAO,qBAAK,IAAI,IAAI;CAChD;;;;;;;;CASA,aAAa,UACX,UACA,SAC4D;EAC5D,MAAM,iCAAiB,IAAI,IAGzB;EACF,IAAI,SAAS,WAAW,GACtB,OAAO;EAIT,MAAM,UAAU,OAAM,MADG,sBAAsB,OAAO,OAAO,EAAA,CAC5B,KAAK,EACpC,OAAO,EAAE,cAAc,SAAS,EAClC,CAAC;EAED,KAAA,MAAW,UAAU,SAAS;GAC5B,IAAI,CAAC,eAAe,IAAI,OAAO,OAAO,GACpC,eAAe,IAAI,OAAO,yBAAS,IAAI,IAAI,CAAC;GAE9C,eAAe,IAAI,OAAO,OAAO,CAAA,EAAG,IAAI,OAAO,QAAQ,OAAO,UAAU;EAC1E;EAEA,OAAO;CACT;;;;;;;;;CAUA,aAAa,QACX,SACA,QACA,SAC8C;EAM9C,QAAO,OAJe,MADG,sBAAsB,OAAO,OAAO,EAAA,CAC5B,KAAK;GACpC,OAAO;IAAE;IAAS;GAAO;GACzB,OAAO;EACT,CAAC,EAAA,CACc,EAAC,EAAG;CACrB;;;;;;;;CASA,aAAa,SACX,MAMA,SACsB;EACtB,MAAM,uBAAuB,iBAAiB,KAAK,UAAU;EAC7D,MAAM,aAAa,MAAM,sBAAsB,OAAO,OAAO;EAG7D,MAAM,kBAAkB,MAAM,WAAW,KAAK;GAC5C,OAAO;IAAE,SAAS,KAAK;IAAS,QAAQ,KAAK;GAAO;GACpD,OAAO;EACT,CAAC;EAED,IAAI,gBAAgB,SAAS,GAAG;GAE9B,MAAM,WAAW,gBAAgB;GACjC,SAAS,aAAa,KAAK;GAC3B,SAAS,aAAa;GACtB,MAAM,SAAS,KAAK;GACpB,OAAO;EACT;EAGA,MAAM,SAAS,MAAM,WAAW,OAAO;GACrC,SAAS,KAAK;GACd,YAAY;GACZ,QAAQ,KAAK;GACb,YAAY,KAAK;GACjB,MAAM,GAAG,KAAK,QAAO,GAAI,KAAK;EAChC,CAAC;EACD,MAAM,OAAO,KAAK;EAClB,OAAO;CACT;AACF;AA5JE,gBAAA,CADC,SAAS,EAAE,UAAU,KAAK,CAAC,CAAA,GALjB,YAMX,WAAA,YAAA,CAAA;AAUA,gBAAA,CADC,MAAM,EAAE,MAAM,OAAO,CAAC,CAAA,GAfZ,YAgBX,WAAA,WAAA,CAAA;AAMA,gBAAA,CADC,MAAM,EAAE,MAAM,OAAO,CAAC,CAAA,GArBZ,YAsBX,WAAA,cAAA,CAAA;AAMA,gBAAA,CADC,MAAM,EAAE,MAAM,OAAO,CAAC,CAAA,GA3BZ,YA4BX,WAAA,UAAA,CAAA;AAWA,gBAAA,CADC,MAAM;CAAE,MAAM;CAAQ,WAAW;AAAK,CAAC,CAAA,GAtC7B,YAuCX,WAAA,cAAA,CAAA;AAMA,gBAAA,CADC,MAAM,EAAE,MAAM,UAAU,CAAC,CAAA,GA5Cf,YA6CX,WAAA,iBAAA,CAAA;AA7CW,cAAN,gBAAA,CAPN,aAAa,EAAE,MAAM,WAAW,CAAC,GACjC,KAAK;CACJ,WAAW;CACX,KAAK,EAAE,SAAS;EAAC;EAAQ;EAAO;EAAU;EAAU;CAAQ,EAAE;CAC9D,KAAK,EAAE,SAAS,CAAC,QAAQ,KAAK,EAAE;CAChC,KAAK;AACP,CAAC,CAAA,GACY,WAAA;AAuKN,IAAM,wBAAN,cAAoC,eAA4B;CACrE,OAAgB,aAAa;;;;;;CAO7B,MAAM,aAAaA,WAA0C;EAC3D,OAAO,KAAK,KAAK,EAAE,OAAO,EAAE,UAAAA,UAAS,EAAE,CAAC;CAC1C;;;;;;;;;;CAWA,MAAM,aAAqC;EACzC,OAAO,YAAyB,IAAI;CACtC;;;;;;;;;;CAWA,MAAM,gBAAgBA,WAA0C;EAC9D,OAAO,iBACL,MACAA,WACA,6BACF;CACF;AACF"}