Pranay Prakash

This commit **fixes a race condition** that caused duplicate `hook_disposed` events to be recorded in the event log when multiple concurrent workflow invocations attempted to dispose of the same hook. To prevent this **event log corruption**, the **`world-local`** storage now employs an exclusive lock file for `hook_disposed` operations, while **`world-postgres`** uses an atomic `DELETE ... RETURNING` query to detect prior disposal. Additionally, the **`core/runtime/suspension-handler`** gains improved logging to differentiate between a hook already being disposed and a run already completed. This **bug fix** significantly enhances the **data integrity** and reliability of hook lifecycle tracking within the system.

This commit introduces **new capabilities** to the **workflow virtual machine (VM) context** by making `btoa`, `atob`, and `Buffer` globals available. This **enhancement** allows scripts executed within the VM to perform **base64 encoding and decoding** and directly manipulate **binary data**. The change primarily affects the `packages/core/src/vm` module, specifically the `createContext` function, and includes new tests to verify the functionality. This significantly expands the types of operations and data handling possible within the VM environment, improving the flexibility of executed workflows.

This commit **improves error handling** within the **workflow runtime** by introducing specific `StepNotRegisteredError` and `WorkflowNotRegisteredError` types. This **bug fix** ensures that when a step or workflow function is not found, the system **fails gracefully** instead of queuing retries or throwing generic errors. A missing step function will now cause the step to fail with `StepNotRegisteredError` (allowing the workflow to catch it), while a missing workflow function will immediately fail the entire run with `WorkflowNotRegisteredError`. This change enhances **system reliability** and provides clearer diagnostics, supported by comprehensive **e2e tests** and updated **documentation**.

This commit **updates the Next.js dependency** to version 16.2.1 across the `docs`, `packages/next`, and several `workbench` environments, including `nextjs-turbopack` and `swc-playground`. This **maintenance update** is coupled with a **bug fix** for the **Next.js deferred build process**. The `packages/next/src/builder-deferred.ts` module now correctly copies all discovered step sources, which is crucial for enabling proper transformations in step-mode builds. This ensures compatibility with the latest Next.js release and resolves a critical build-time defect affecting development workflows.

This commit performs a **dependency upgrade** for the `nitropack` library within the **`workbench/nitro-v2`** environment, updating it from version `^2.12.7` to `^2.13.1`. This **maintenance chore** ensures the project benefits from the latest features, bug fixes, and performance enhancements provided by `nitropack`. The update primarily impacts the **server-side rendering and API routing capabilities** of the `nitro-v2` workbench, aligning it with recent upstream improvements. A new changeset file was also added to reflect this patch update for workflow packages.

This commit performs a **dependency update** across the project's Nuxt ecosystem, specifically **bumping the versions** of several key packages. It updates `@nuxt/kit`, `@nuxt/schema`, and `nuxt` within the `packages/nuxt/package.json` file, affecting the core Nuxt package. Additionally, the `workbench/nuxt/package.json` also sees an update to its `nuxt` dependency, ensuring the development environment is current. This **maintenance chore** ensures the project remains compatible with the latest Nuxt releases and benefits from upstream improvements and bug fixes.

This commit performs a **maintenance chore** by **updating dependencies** within the **SvelteKit workbench**. It specifically bumps the versions of core SvelteKit ecosystem packages, including `@sveltejs/kit`, `svelte`, `@sveltejs/adapter-node`, and `@sveltejs/adapter-vercel`, as reflected in `workbench/sveltekit/package.json`. This ensures the `sveltekit` workbench benefits from the latest features, bug fixes, and performance improvements provided by these updated libraries, enhancing its stability and compatibility with the broader Svelte ecosystem.

This commit performs **dependency updates** within the project's **workbench examples**. Specifically, it **bumps the `express` dependency** from `^5.1.0` to `^5.2.1` in both the `workbench/express/package.json` and `workbench/nest/package.json` files. Additionally, the `fastify` dependency is updated from `^5.6.2` to `^5.8.2` in `workbench/fastify/package.json`. This **maintenance chore** ensures that the example workbenches are utilizing more recent and potentially more stable or secure versions of these popular web frameworks.

This commit performs a **maintenance update** by bumping the versions of several **NestJS core dependencies**. Specifically, it updates `@nestjs/common`, `@nestjs/core`, and `@nestjs/platform-express` to **`^11.1.17`** within both the `packages/nest` and `workbench/nest` modules. This ensures that these parts of the application leverage the latest features, bug fixes, and performance enhancements provided by the updated NestJS framework. The change helps maintain compatibility and keeps the project current with the NestJS ecosystem.

This commit performs a **dependency update** for the **`hono` web framework** within the `workbench/hono` module. It specifically bumps the `hono` package version from `^4.9.10` to `^4.12.8` in the `workbench/hono/package.json` file. This **maintenance chore** ensures that the `workbench`'s `hono`-based components leverage the latest features, bug fixes, and performance improvements from the updated framework version, enhancing overall stability and functionality.

This commit **reverts** a previous change that inlined class serialization registration, restoring the mechanism where the **SWC plugin** explicitly imports and calls the `registerSerializationClass` function from `@temporalio/core`. This **refactoring** impacts the core serialization utilities and the `packages/swc-plugin-workflow` transformation logic, specifically in `transform/src/lib.rs`. The change ensures that class serialization is handled by importing the `registerSerializationClass` function, rather than embedding its logic directly into the transformed code. This **maintenance** task addresses potential issues with 3rd-party package support and updates numerous test outputs to reflect the revised code generation.

This commit introduces a **new public API for semantic error types** under `workflow/errors`, allowing consumers of the `@workflow` package to import and handle specific errors like `HookNotFoundError` and `EntityConflictError`. It **refactors the error hierarchy** within `packages/errors`, standardizing the `retryAfter` property on `TooEarlyError` and `ThrottleError` to be a number of seconds for consistency with `WorkflowWorldError`, which required **bug fixes** in several `world-*` packages. Additionally, comprehensive **API reference documentation** for these error types has been added and reorganized into a dedicated `api-reference/workflow-errors` section. This enhances error handling capabilities and improves developer experience by providing clear, accessible error definitions.

This commit introduces a **new feature** to enhance the **reliability and error handling** of workflow and step execution by enforcing a maximum number of queue deliveries. It shifts the responsibility for capping retries from the Vercel Queue Service (VQS) configuration to the **core runtime handlers**, specifically `workflowEntrypoint` and `stepHandler`. After `MAX_QUEUE_DELIVERIES` (now 48) attempts, a run or step will **gracefully fail** with a new `MAX_DELIVERIES_EXCEEDED` error, preventing "phantom stuck" states. Additionally, the **local queue implementation** (`createQueue`) is improved with a safety limit and linear backoff to better approximate VQS retry behavior, ensuring consistent failure handling across environments. This change provides clearer failure states for excessively retried messages and improves overall system resilience.

This commit introduces a **major new capability** by enabling the `start()` function to be called directly within **workflow functions**, allowing for dynamic creation and management of child workflows from within a running workflow. It achieves this by routing `start()` calls through an internal step and introducing a `WorkflowRun` class with step-backed methods (e.g., `cancel()`, `status`, `returnValue`) for interacting with child `Run` objects, supported by custom **serialization** logic. This significantly enhances **workflow composition** and **inter-workflow communication**, enabling patterns like recursive and repeating workflows. Furthermore, it improves **observability** by rendering `Run` objects as clickable links in the web UI, and includes extensive **documentation** updates and new **e2e tests** to validate these features.

This commit significantly **enhances the `workflow` skill** by introducing comprehensive **testing documentation** within `skills/workflow/SKILL.md`. This **new feature** provides detailed guidance on best practices for unit, integration, and webhook testing, complete with illustrative code examples. The update aims to standardize and improve the quality of testing procedures for workflow development, ultimately leading to more robust and reliable workflow implementations.

This commit delivers a **documentation update** for the `workflow` skill, advancing the documented **workflow version from 1.3 to 1.4**. A new section has been added to `skills/workflow/SKILL.md` that thoroughly explains the use of **namespaced streams**. This **feature enhancement** provides users with practical code examples for separating `log levels` and `agent output`, significantly improving the organization and clarity of workflow execution. The update empowers users to better manage and understand complex workflow processes by leveraging these new stream capabilities.

This commit introduces a comprehensive suite of **new stress benchmarks** to evaluate the performance and resilience of the **workflow execution engine** under heavy load. It adds new benchmark suites in `packages/core/e2e/bench.bench.ts` for **1000-step sequential workflows**, **large data payload processing**, and **complex stream manipulation**, supported by new example workflow definitions in `workbench/example/workflows/97_bench.ts` such as `sequentialDataPayloadWorkflow` and `streamPipelineWorkflow`. This **enhances the project's testing capabilities** by simulating high-load scenarios and integrates these new tests into the CI/CD benchmarking pipeline through updates to `.github/workflows/benchmarks.yml`. The work ensures better performance monitoring and stability assurance for critical system components by providing robust stress testing coverage.

This commit introduces **new documentation** for **error codes and retry mechanisms** within the `errors-and-retries` guide, providing users with a comprehensive understanding of workflow failure causes and programmatic access to error details. Concurrently, it performs a significant **refactoring and bug fix** by **removing the import inference system** from the `packages/docs-typecheck` module. This system previously masked issues by silently adding imports to documentation snippets, which could hide incorrect or missing dependencies. As a result, all 24 affected documentation code snippets now include explicit and correct imports, ensuring their accuracy and preventing future silent failures. The overall impact is enhanced user guidance and a more robust documentation build process that accurately reflects runnable code examples.

This commit introduces a **new `HookConflictError` class** within the **error handling subsystem** to specifically address conflicts arising from duplicate hook tokens. This **refactoring** replaces the generic `WorkflowRuntimeError` for these scenarios, ensuring that such conflicts are correctly classified as a `USER_ERROR` rather than a runtime issue. The change improves **error specificity** for developers interacting with **workflow hooks**, providing clearer feedback when a user attempts to register an already-used token. Additionally, documentation is updated to leverage the `HookConflictError.is()` static method for robust cross-VM error checking, enhancing the overall reliability and clarity of error management.

This commit significantly **enhances the documentation** for the **`@workflow/world` package's API**, focusing on workflow steps and runs. It adds **TSDoc** to the `stepName` and `workflowName` fields within `StepSchema` and `WorkflowRunBaseSchema`, explaining their machine-readable format and guiding users to the `parseStepName()` and `parseWorkflowName()` utilities for extraction. Furthermore, the **`world.steps` API reference documentation** is expanded with practical examples, demonstrating efficient step listing using `resolveData: 'none'` and data hydration with `observabilityRevivers`. This **documentation enhancement** improves developer clarity and usability when interacting with workflow data and parsing step/workflow names for display.