feat: Queue/Network refactor + SDK warmup #212
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Implements SQLite-based event queue to solve WebSocket message loss (Issue #205). Components: - @agentxjs/types/queue - EventQueue interface and protocol types - @agentxjs/queue - SQLite implementation with cursor support - createLocalAgentX - Queue integration (subscribe/ack handlers) Features: - Cursor-based reconnection recovery - Topic partitioning by sessionId - ACK marking for cleanup - Bun/Node.js 22+ auto-detection Verified: - Unit tests pass (7/7) - Integration test confirms recovery works TODO: - Browser client ACK implementation - Auto cleanup of acknowledged entries Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Add channel abstraction to WebSocketServer for targeted message routing. Changes: - ChannelServer interface: subscribe/publish/unsubscribe methods - WebSocketServer: channel subscription management - Auto cleanup: connections unsubscribed on close - README: comprehensive API documentation Features: - Multi-channel subscription per connection - Channel isolation (publish only to subscribers) - Idempotent subscription (Set-based) - Auto cleanup on disconnect Verified: - Channel isolation works - Multi-subscriber support - Auto cleanup on close Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Implement standard MQ pattern with multiple independent consumers. Changes: - EventQueue API: add consumerId to read/ack/subscribe - New methods: createConsumer, getConsumerCursor, deleteConsumer - Database: add queue_subscriptions table for consumer tracking - Cleanup: delete entries consumed by all consumers (MIN cursor) - Remove acknowledged fields from queue_entries (moved to subscriptions) Architecture: - Each consumer independently tracks consumption progress - Broadcast semantics: all consumers receive all messages - Cleanup based on MIN(consumer cursors) across all consumers - Orphaned entries (no consumers) cleaned by timestamp TTL Verified: - Multi-consumer independence (8/8 tests pass) - ACK updates consumer cursor correctly - Cleanup respects MIN cursor across consumers Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Phase 2 Architecture Changes: - Remove Channel capabilities from Network (subscribe/publish/unsubscribe) - Add MessageSender interface for transport abstraction - Implement handleConnection() for protocol encapsulation - Add dual TTL cleanup (Consumer 24h + Message 48h) - Simplify AgentX server (~265 → ~170 lines) - Add auto-subscribe/ACK/reconnect in AgentX client BDD Test Redesign: - Remove outdated feature files (old Container/Image/Agent API) - Add Layer 1 features: local-mode, remote-mode, conversation lifecycle - Add Layer 2 features: reconnect, multi-consumer, delivery guarantees - Update world.ts for new session-based model Documentation: - Update issue 046 with Phase 2 completion status - Add issue 047 for unified development mode (Code Review + BDD) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Step Definitions: - agentx.steps.ts: Local/remote mode, server operations, event subscription - conversation.steps.ts: Container, Image, Agent, Message lifecycle - reliability.steps.ts: Reconnection, multi-consumer, delivery guarantees Infrastructure: - Update world.ts with additional state tracking - Update cucumber.js with new profiles (local, remote, reliability, etc.) - Update package.json scripts for new test profiles Test Commands: - bun test (Layer 1 basic) - bun test:reliability (Layer 2) - bun test:integration (Claude API) - bun test:all (everything) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Test Infrastructure: - Add test-server.ts: Dedicated WebSocket server for tests (port 15300) - Add test-manager.ts: Unified test runner like dev-manager - Simplify cucumber.js: Single default config, use tags for filtering - Update package.json: Add "test:bdd" command Step Definitions: - Replace bun:test with Node.js assert for Cucumber compatibility - Add expect() helper functions in all step files - Simplify local-mode.feature to 3 basic scenarios Usage: bun test:bdd # Start server + run tests bun test:bdd --tags @Local # Run specific tests bun test:bdd server # Start server only bun test:bdd run # Run tests only Results: 3 scenarios (3 passed), 10 steps (10 passed) in 0.144s Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Changes: - Remove Background server setup in all features (use global test-server) - Unify all ports to 15300 - Add bdd/.env.local (copied from dev) and .env.example - Update test-server to load bdd/.env.local - Remove duplicate step definitions (keep in conversation.steps.ts) - Fix dispose() to set isConnected = false Test Results: 10/11 scenarios pass - 4/4 local mode ✅ - 6/7 remote mode (1 timeout, investigating) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Changes: - Add toBeUndefined() to conversation.steps.ts expect helper - Add empty object request step for container_list_request - Fix isConnected state in dispose step Test Results: - Local mode: 4/4 ✅ - Container: 4/4 ✅ - Remote: 6/7 (1 request timeout, investigating) - Image: 4/5 (data isolation issue, need unique containers) Overall: 23/52 scenarios pass (27 undefined, 2 failed) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Added Steps: - Global test server setup step (port 15300) - Client connection and subscription steps - Message sending variants (rapid succession, multiple clients) - Consumer lifecycle (disconnect, reconnect, cleanup) - Cross-device and multi-topic steps - Additional client state tracking steps Fixes: - Update all old port references (15220/15230/15240) to 15300 - Add .env.local and .gitignore to bdd/ Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Changes:
- Add pendingSubscriptions Map to track subscription promises
- Make subscribeToTopic() return Promise<void>
- Resolve promise when queue_subscribed confirmation received
- Await subscribeToTopic("global") before returning AgentX instance
This ensures remote clients receive responses for global commands
(container_create, image_create, etc.) that don't have sessionId.
BDD: Add missing reliability step definitions for multi-consumer tests
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Critical Bug Fix: When handling queue_entry messages, extract the inner event and check if it's a response to a pending request BEFORE dispatching to handlers. Previously, queue_entry handler would dispatch to event handlers and return immediately, never reaching the pending request resolution logic. This caused all remote client requests to timeout even though responses were delivered via the queue. Changes: - Move pending request check into queue_entry handler - Add "Resolving pending request from queue" log - Return early after resolving (don't dispatch to handlers) Test Results: - Remote mode: 7/7 ✅ (was 6/7) - Layer 1 total: 24/25 ✅ (was 13/25) This fix is essential for Queue-based reliable delivery to work. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Proposes a Mock LLM Driver for BDD tests to enable: - Fast, predictable testing without API calls - Complete event flow validation (Agent → Queue → Client) - Layer 2 reliability tests with real message streaming - Zero cost, deterministic scenarios Design includes: - LLMDriver interface abstraction - MockClaudeDriver implementation in bdd/mock/ - Injection mechanism via llmProvider config - Predefined scenarios (simple-text, with-tool, error, etc.) Next: Research Driver architecture and implement injection point Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Major Update: - AgentX uses Environment pattern (not Driver pattern) - Environment = Receptor (SDK→Bus) + Effector (Bus→SDK) - Each Agent creates its own ClaudeEnvironment Design Changes: - Propose EnvironmentFactory for dependency injection - MockEnvironment implements existing Environment interface - MockEffector listens to user_message, emits predefined events - MockReceptor emits mock events to SystemBus Benefits: - Reuses existing abstractions (no new interfaces needed) - Clean injection via factory pattern - Mock implementation isolated in bdd/mock/ - Enables true end-to-end testing (Agent → Queue → Client) Next: Implement Phase 1 (add environmentFactory to RuntimeConfig) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Phase 1: Architecture Changes New Types: - EnvironmentFactory interface (packages/types/src/runtime/internal/) - EnvironmentCreateConfig interface - Export from runtime/internal Implementation: - DefaultEnvironmentFactory (packages/runtime/src/environment/) - Wraps ClaudeEnvironment creation with factory pattern Integration via Context Pattern: - RuntimeConfig.environmentFactory (optional) - RuntimeContainerContext.environmentFactory (passed via context) - RuntimeAgentConfig.environmentFactory (received from context) - RuntimeAgent uses factory ?? defaultEnvironmentFactory Configuration Flow: LocalConfig → createRuntime → RuntimeImpl → createContainerContext → RuntimeContainer → RuntimeAgent Benefits: - No layer-by-layer parameter passing (uses context) - Default behavior unchanged (defaultEnvironmentFactory) - Enables MockEnvironment injection for BDD tests Test: bun test:bdd --tags @Local (4/4 passed) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Phase 2: Mock Implementation
Files Created (bdd/mock/):
- MockReceptor.ts: Emits events to SystemBus
- MockEffector.ts: Listens to user_message, emits predefined events
- MockEnvironment.ts: Combines Receptor + Effector
- MockEnvironmentFactory.ts: Factory with scenario control
- scenarios.ts: 7 predefined scenarios
- index.ts: Public exports
Scenarios:
- default: Simple text response
- multi-delta: Multiple text chunks
- with-thinking: Thinking blocks
- with-tool: Tool use
- error: Error scenario
- long-stream: 100 text deltas (for reliability tests)
- instant: No delays
Usage:
```typescript
const factory = new MockEnvironmentFactory();
factory.setScenario("long-stream");
const agentx = await createAgentX({
environmentFactory: factory,
});
```
Next: Phase 3 (BDD integration)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Phase 3: BDD Integration Changes: - Fix RuntimeImpl: Store environmentFactory field - Fix createContainerContext: Use this.environmentFactory - Update world.ts: Add createMockAgentX() and setMockScenario() - Add mock.steps.ts: Steps for mock environment control - Add message-flow.feature: 4 @mock test scenarios Features: - Mock text responses (instant, multi-delta, long-stream) - Event order verification - Text content validation Test Results: - 38/53 scenarios pass (up from 31/52) - Mock tests run successfully (need debugging for duplicate messages) Known Issue: - Text appears twice ("Instant responseInstant response") - Investigating MockEffector event emission Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Bug Fix: - MockEffector was emitting broadcastable: true events - This caused events to be delivered twice: 1. Original MockEffector event → agentx.on() 2. BusPresenter processed event → agentx.on() Solution: - Set broadcastable: false in MockEffector (like ClaudeReceptor) - Filter broadcastable: false in LocalAgentX.on() - Add wait logic in mock.steps.ts for async event delivery Test Results: - Mock tests: 4/4 ✅ - Overall: 42/56 (75%) ⬆️ from 38/53 - Steps: 353/389 (91%) Added: - debug-mock.ts for testing MockEnvironment - Wait loop in "I should receive exactly N events" step Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Major Refactoring: Replace broadcastable field with semantic event classification based on source and intent fields. Changes: - Add shouldEnqueue() helper: Filters by source/intent instead of broadcastable - source: "environment" → Internal (DriveableEvent) - intent: "request" → Internal (user_message, interrupt) - All others → External (enqueue) - Update createLocalAgentX.on(): Filter using shouldEnqueue() - Remove all broadcastable: false assignments - Update comments to use source-based terminology Files Modified: - packages/agentx/src/createLocalAgentX.ts - packages/runtime/src/environment/ClaudeReceptor.ts - packages/runtime/src/internal/AgentInteractor.ts - packages/runtime/src/internal/RuntimeAgent.ts - bdd/mock/MockEffector.ts Benefits: - Clearer semantics (source/intent vs boolean flag) - Single source of truth for event classification - No duplicate filtering logic - Prepares for eventual removal of broadcastable field Test Results: 42/56 (75%) - unchanged, refactoring successful Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Data Isolation Strategy: - All tests use bdd/.agentx-test/ instead of ~/.agentx - test-manager cleans .agentx-test/ before each run - Each scenario gets unique container ID (scenarioId_containerName) - Container name mapping stored in world.savedValues Changes: - test-manager.ts: Add cleanTestData() before server/tests - bdd/.gitignore: Add .agentx-test/ - world.ts: Add scenarioId field (unique per scenario) - agentx.steps.ts: Use .agentx-test for local instances - conversation.steps.ts: - Generate unique container IDs - Resolve container name → unique ID in all steps - Add debug output for image list failures Test Results: - @image: 5/5 ✅ (was 4/5) - Overall: 42/56 (75%) - Data isolation issue resolved Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
…lity scenarios Changes: - test-server.ts: Support MOCK_LLM env var and MockEnvironmentFactory - scenarios.ts: Add ordered-messages, slow-stream for reliability tests - queue-delivery.feature: 4 new scenarios using real message flow - mock.steps.ts: Add When import and reliability step definitions New Scenarios: - Messages delivered in order through Queue - Multiple clients receive same stream independently - Disconnect during streaming recovers all messages - Queue handles long event stream (100 events) Steps Added: - Server mock scenario control - Remote client setup with named clients - Text delta subscription and verification - Disconnect/reconnect with resubscription - Event counting and ordering validation Next: Test and debug new scenarios Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Changes: - Simplify queue-delivery.feature scenarios (3 tests) - Remove dependency on dynamic mock scenario switching - Add new step definitions for "at least N events" assertions - Focus on verifying Queue message delivery, not scenario control Scenarios: 1. Messages delivered through Queue 2. Multiple clients receive same stream independently 3. Client receives messages through Queue All use default MockEnvironment scenario for simplicity. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
New Files: - features/integration/real-api.feature: 2 @integration scenarios - steps/integration.steps.ts: Real API test steps Scenarios: 1. Capture real API event flow (records events to JSON) 2. Disconnect during real API streaming recovers messages Purpose: - Test core Queue功能 with real Claude API - Capture actual event flows to improve Mock scenarios - Verify disconnect/reconnect message recovery Usage: MOCK_LLM=false bun test:bdd --tags "@capture-events" Output: bdd/mock/captured-scenario.json (real API event sequence) Next: Run integration tests and improve Mock based on real events Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Documents remaining work for complete Queue MQ validation: Phase 1 (Must): Fix integration tests - Fix step definition conflicts - Debug cucumber @integration tag filtering - Run real API tests Phase 2 (Recommended): Capture real events - Run @capture-events scenario - Generate captured-scenario.json from real API - Analyze event types, order, data structure Phase 3 (Must): Validate core scenario - Run @disconnect-recovery test - Verify: WebSocket disconnect → 0% message loss (was 35.7%) - Verify: Reconnect → Queue recovers all missed events Phase 4 (Optional): Improve Mock - Update scenarios.ts based on real API events - Rewrite @Reliability tests using accurate Mock Current Status: - Queue architecture: Complete ✅ - Unit tests: 9/9 pass ✅ - BDD framework: Complete ✅ - Layer 1 tests: 29/29 pass ✅ - Core scenario: Not yet validated⚠️ Next: Fix integration tests and run Phase 1-3 Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
…reate - Add await to subscribeToTopic in createAgentX.ts to fix race condition - Add baseUrl support to test-server.ts for relay API - Create cucumber.integration.js config for integration tests - Increase default timeout to 120s for real API calls - Rewrite integration tests to use remote client mode - Add comprehensive step definitions for disconnect recovery testing - Capture real API event flow to mock/captured-scenario.json Integration tests now validate: - Real API event capture (@capture-events) - Message recovery after disconnect (@disconnect-recovery) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Previously, events were persisted to SQLite before notifying subscribers, causing streaming to be blocked by DB write latency. Multiple concurrent writes would serialize, resulting in "bursting" behavior where events arrive in batches rather than streaming smoothly. Changes: - Reorder append() to notify subscribers FIRST, then persist to DB - Add WAL mode for better concurrent read/write performance This ensures events are delivered immediately without waiting for DB, while still maintaining persistence for reconnection recovery. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
…gging - Add persistenceEnabled option to skip DB writes (default: true) - Add queueEnabled option for pure pass-through mode (default: true) - Useful for isolating latency issues in Queue vs other components DEBUG: Currently set queueEnabled: false in createLocalAgentX for testing Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add localStorage debounce for cursor storage (flush every 500ms) - Fix session ID capture to only save once per session - Add BYPASS_QUEUE debug flag for testing direct WebSocket broadcast - Remove unused @ts-expect-error directives in SqliteEventQueue and sqlite driver - Fix import paths in DefaultEnvironmentFactory - Add dispose() method to Environment interface - Fix McpServerConfig type in EnvironmentFactory Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Move message persistence from BusPresenter to Queue ACK callback. This ensures messages are only persisted after client acknowledges receipt, keeping Session and Queue cursor synchronized. Changes: - Remove direct session.addMessage from BusPresenter - Add onAck callback option to QueueOptions - Add getEntryByCursor method to SqliteEventQueue - Implement ACK-driven persistence in createLocalAgentX This fixes the synchronization issue where Session could have messages that the client never received (due to disconnect), causing duplicate or missing messages on reconnect. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Reorganize network package for better maintainability: - protocol/reliable-message.ts: ACK protocol types and guards - server/WebSocketConnection.ts: Connection wrapper with heartbeat and sendReliable - server/WebSocketServer.ts: Server management (listen, attach, broadcast) - client/WebSocketClient.ts: Node.js client with auto-ACK - client/BrowserWebSocketClient.ts: Browser client with reconnect and auto-ACK - factory.ts: Environment-aware client factory Each file now has a single responsibility and reasonable size (30-230 lines). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add comprehensive tests for network package: - protocol/reliable-message.test.ts: Type guard tests - server/WebSocketConnection.test.ts: Connection, sendReliable, ACK handling - server/WebSocketServer.test.ts: Server lifecycle, broadcast, integration 42 tests covering: - isAckMessage and isReliableWrapper type guards - send and sendReliable message delivery - ACK callback and timeout handling - Server listen, broadcast, close - sendReliable integration with auto-ACK Also fix server close timeout to prevent test hanging. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add WebSocketClient tests (10 tests): - Connection lifecycle - Send/receive messages - Auto-ACK for reliable messages - Event handlers - Rewrite README to reflect current architecture: - Remove outdated Channel pub/sub docs - Document sendReliable and ACK protocol - Add server and client API reference - Include architecture diagram - Add usage examples Total: 52 tests passing Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Extract createRemoteAgentX to separate file - Simplify createAgentX.ts to pure factory (~45 lines) - Use Network layer auto-ACK instead of manual queue_* messages - Remove cursor storage and clientId generation - Simplify subscription to just "subscribe" message Code reduction: ~500 lines → ~310 lines total Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Major refactoring of the queue system: **New Architecture**: - In-memory pub/sub using RxJS Subject (real-time, fast) - SQLite persistence via native APIs (async, non-blocking) - Decoupled from network protocol (caller decides when to ACK) **common/sqlite**: New unified SQLite abstraction - Auto-detects runtime: Bun (bun:sqlite) or Node.js 22+ (node:sqlite) - Zero external dependencies (removed db0) - Exported via `@agentxjs/common/sqlite` **New EventQueue API**: - `publish(topic, event)` - broadcast + async persist - `subscribe(topic, handler)` - real-time subscription - `ack(consumerId, topic, cursor)` - update consumer position - `getCursor(consumerId, topic)` - get last ACKed position - `recover(topic, afterCursor?)` - fetch historical events - Removed: createConsumer, read, deleteConsumer, handleConnection **Updated packages**: - @agentxjs/common: Add sqlite module with separate build target - @agentxjs/queue: Rewrite with RxJS, remove db0 - @agentxjs/persistence: Use common/sqlite, remove db0 peer dep - agentxjs: Update createLocalAgentX for new Queue API Tests: 13 queue tests, 52 network tests passing Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add @agentxjs/common/path for cross-runtime path utilities: - getModuleDir(import.meta) - current module directory - getPackageRoot(import.meta) - package root (package.json) - getMonorepoRoot(import.meta) - monorepo root (lock files) - resolveFromRoot/resolveFromPackage - path resolution helpers Works in both Bun and Node.js environments. Also: - Auto-create parent directories in openDatabase() - Fix BDD tests to use new path utilities instead of import.meta.dir - Add 12 unit tests for path module Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add comprehensive documentation for path module: - getModuleDir - cross-runtime __dirname - getPackageRoot - find package.json - getMonorepoRoot - find workspace root - resolveFromRoot/resolveFromPackage - path helpers Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Add documentation for new tools to guide AI assistants: - common/sqlite: Native SQLite abstraction (Bun/Node.js) - common/path: Cross-runtime path utilities - queue: New RxJS-based API (publish/subscribe/ack) - network: sendReliable() for at-least-once delivery This ensures future AI assistants use the correct tools. Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Replace async persistence with sync writes: - Remove pendingPersists tracking complexity - Remove setImmediate async logic - Use sync SQLite writes (fast, microseconds) - Add isClosed guard to prevent writes after close Fixes "database is not open" errors in BDD tests. Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Update bdd test configuration: - Use test-manager for `bun run test` (auto-starts server) - Add test:server and test:run scripts - Fix import.meta.dir usage with common/path Test improvements: - 43 scenarios passing (up from 21) - 370 steps passing (up from 163) - Server auto-management working Remaining: 9 failures, 7 undefined (mostly business logic) Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Update step definitions to handle server vs local mode: - Check for usedPorts to determine if server is running - Use remote client when server exists (avoid port conflict) - Use local mode when no server - Fix containerIds assertion to handle test prefixes - Add message collection on assistant_message events - Implement real message_send instead of simulation Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
**Critical fix**: Correct dispose() order to prevent zombie processes - OLD: queue → wsServer → runtime (runtime still emits after queue closed) - NEW: wsServer → runtime → queue (safe shutdown order) **test-server improvements**: - Add dispose timeout protection (5s) - Improve shutdown signal handling **test-manager improvements**: - Add graceful shutdown with SIGTERM - Force kill with SIGKILL after 5s timeout - Wait for process exit properly **BDD shutdown tests** (4 scenarios, 21 steps): - Local mode dispose < 2s ✅ - Dispose idempotent ✅ - Server shutdown < 2s ✅ - Multiple clients < 2s ✅ Fixes zombie process issue (CPU 97% hangs). Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Replace simulation with real message_send_request calls: - server sends message → real message_send_request - server sends N messages → loop with real sends - Proper imageId lookup from createdImages - Add error handling for missing images Also: - Subscribe to assistant_message to collect received messages - Fix image lookup fallback (throw instead of using alias) Progress: More reliability tests now execute real message flows. Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Add AgentXResponse as the base type for all command responses, providing: - Unified structure: requestId, error, __subscriptions - Auto-subscription: client automatically subscribes to sessions - Type safety: createResponse() now requires D extends AgentXResponse This fixes the "Queued..." stuck issue after reconnection by ensuring clients auto-subscribe to all returned sessionIds. Changes: - Add AgentXResponse type with hasSubscriptions/isErrorResponse guards - Add generateRequestId() utility to @agentxjs/common - Update all response types to extend AgentXResponse - Make record field optional in ImageCreateResponse/ImageUpdateResponse - Use unified error checking with isErrorResponse() in UI hooks - Server adds __subscriptions to image_create/list/get responses Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add comprehensive README for queue package: - Quick start guide - API reference - Multi-consumer example - Reconnection recovery example Add changeset for minor version bump: - Queue: RxJS + native SQLite refactor - Network: sendReliable() ACK mechanism - Common: sqlite and path utilities - AgentX: protocol simplification Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
Move typecheck to CI to speed up push (~10s faster). Pre-commit still does format/lint on staged files. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add warmup() method to pre-initialize Claude SDK subprocess before the first user message. This reduces cold start latency by starting the SDK early when the Agent is created. Changes: - SDKQueryLifecycle: add warmup() method - ClaudeEffector: expose warmup() method - Environment interface: add optional warmup() method - ClaudeEnvironment: implement warmup() - RuntimeAgent: call warmup() on construction (fire-and-forget) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Set up UI hook testing infrastructure: - Add happy-dom for React hook testing in Bun - Add bunfig.toml for test preload configuration - Add useAgent.test.tsx with event filtering tests Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Major refactoring of Queue and Network layers with new SDK warmup feature.
Queue Package (New Architecture)
Network Package
sendReliable()for at-least-once deliveryRuntime Package
Common Package
@agentxjs/common/sqlite- unified SQLite abstraction@agentxjs/common/path- cross-runtime path utilitiesUI Package
Test plan
🤖 Generated with Claude Code