Release 0.14.0

Changed

• Executor Schema Mechanism — Pydantic BaseModel with ClassVar

• All built-in executors now declare input/output schemas as Pydantic BaseModel classes assigned via inputs_schema: ClassVar[type[BaseModel]] and outputs_schema: ClassVar[type[BaseModel]], replacing inline get_input_schema() / get_output_schema() method overrides

• BaseTask.get_input_schema() and get_output_schema() are now implemented in the base class: they call model_json_schema() on the Pydantic model and resolve $ref references inline via resolve_schema_refs()

• Subclasses no longer need to override get_input_schema() / get_output_schema() — just set the ClassVar

• Both Pydantic BaseModel and legacy dict schemas are still supported (backward compatible)

• Added resolve_schema_refs() helper in core/utils/helpers.py to inline $defs/$ref from Pydantic-generated JSON Schema

• Updated all 16 built-in executors: AggregateResultsExecutor, ApiExecutor, BatchCrewaiExecutor, CrewaiExecutor, DockerExecutor, SendEmailExecutor, GenerateExecutor, GrpcExecutor, RestExecutor, LLMExecutor, McpExecutor, ScrapeExecutor, SshExecutor, CommandExecutor, SystemInfoExecutor, WebSocketExecutor

• Updated documentation across 8 files: extending.md, custom-tasks.md, quick-reference.md, python.md, basic_task.md, quick-start.md, faq.md, best-practices.md

• API Architecture Refactoring — Three-Layer Design

• Introduced unified capabilities registry (api/capabilities.py) as single source of truth for all 15 task operations, replacing separate inline definitions in A2A and MCP modules

• A2A POST / now only handles agent-level actions (tasks.execute, tasks.generate, tasks.cancel); CRUD operations return an error directing clients to POST /tasks

• MCP adapter auto-generates all 15 tools from the capabilities registry via call_by_tool_name routing

• Simplified A2A adapter: removed ~300 lines of CRUD bridging code from agent_executor.py and task_routes_adapter.py

Added

• Email Executor (send_email_executor)

• Send emails via Resend HTTP API or SMTP protocol

• Provider dispatch: inputs["provider"] selects "resend" or "smtp"

• Resend provider: POST to https://api.resend.com/emails via httpx.AsyncClient, returns message ID and status code

• SMTP provider: Uses aiosmtplib.send() (optional dependency, guarded import)

• Supports plain text and HTML emails, cc/bcc recipients, reply-to address

• Input validation for required fields (provider, to, subject, from_email, body/html)

• Schema methods (get_input_schema, get_output_schema) and demo mode support

• Added [email] optional dependency with aiosmtplib>=3.0.0

• Added email to [all] installation extra

• Comprehensive test coverage: 26 tests for validation, Resend, SMTP, schemas, and demo results

• Method Discovery Endpoint (GET /tasks/methods)

• Returns all available task methods grouped by category (agent_action, crud, query, monitoring) with input schemas, descriptions, and examples

• Driven by the capabilities registry for automatic consistency

• Task Scheduling System

• Added ScheduleType enum supporting six schedule types: once, interval, cron, daily, weekly, monthly

• Added 9 scheduling fields to TaskModel: schedule_type, schedule_expression, schedule_enabled, schedule_start_at, schedule_end_at, next_run_at, last_run_at, max_runs, run_count

• Created ScheduleCalculator module for next execution time calculation across all schedule types

• Database migration 002_add_scheduling_fields.py for adding scheduling columns and indexes

• Automatic execution mode detection: tasks with children execute in tree mode, tasks without children execute in single mode

• Internal Scheduler

• Built-in asyncio-based scheduler (InternalScheduler) for standalone deployment

• Configurable poll interval, max concurrent tasks, task timeout, and user filtering

• Scheduler lifecycle management: start, stop, pause, resume

• Task completion callbacks for monitoring and integration

• External Scheduler Gateway

• WebhookGateway for HTTP webhook integration with external schedulers

• HMAC signature validation, IP filtering, and rate limiting

• Helper functions for generating cron entries and Kubernetes CronJob manifests

• Support for cron, Kubernetes CronJob, APScheduler, Temporal integration

• Calendar Integration

• ICalExporter for exporting scheduled tasks in iCalendar format

• RRULE support for recurring events (daily, weekly, monthly, interval)

• Calendar feed URL generation for subscription in Google Calendar, Outlook, etc.

• Scheduler CLI Commands

• apflow scheduler start - Start internal scheduler (foreground or background)

• apflow scheduler stop - Stop running scheduler

• apflow scheduler status - Show scheduler status

• apflow scheduler trigger - Manually trigger a task

• apflow scheduler export-ical - Export tasks to iCal format

• apflow scheduler webhook-url - Generate webhook URL for external schedulers

• Scheduling API Endpoints

• tasks.scheduled.list - List all scheduled tasks

• tasks.scheduled.due - Get tasks due for execution

• tasks.scheduled.init - Initialize/recalculate next_run_at

• tasks.scheduled.complete - Complete scheduled run and calculate next execution

• tasks.webhook.trigger - Trigger task execution via webhook (JSON-RPC)

• POST /webhook/trigger/{task_id} - REST endpoint for external schedulers

• Repository Methods for Scheduling

• get_due_scheduled_tasks() - Query tasks ready for execution

• get_scheduled_tasks() - List scheduled tasks with filters

• mark_scheduled_task_running() - Mark task as in-progress

• complete_scheduled_run() - Complete run and update scheduling state

• initialize_schedule() - Calculate initial next_run_at

• Added scheduling optional dependency with croniter>=1.0.0

• Added scheduling to standard and all installation extras

• Comprehensive test coverage: 118 tests for scheduling functionality