Verifiable coding agent runtime
Defines the runtime, loop, workflow, and software-factory layers, then explains what the runtime controls.
Atomic resources
Use this index to find the right Atomic reference without reading the same material twice. Product explainers define concepts. Maintained documentation specifies current behavior. Source links expose the implementation. Selected field notes record the engineering context behind particular workflow designs.
Where should I start?
Start with the runtime explainer if Atomic is new to you. Use the TypeScript SDK guide when you are ready to author a workflow. Use the built-in workflow catalog to compare shipped control and review shapes. Use the hosted documentation for the complete current API.
Evergreen Atomic guides
Coding agent workflows for agentic engineering
A conceptual guide to stages, handoffs, dynamic execution, checks, artifacts, review, and recovery conditions.
Verification for coding agents
Provides a practical model for acceptance contracts, evidence, fresh-context review, gates, retries, checkpoints, and human escalation.
Workflow SDK for TypeScript
Provides current authoring syntax and verified examples for contracts, tasks, parallel branches, artifacts, human input, and composition.
Built-in workflow examples
Compares selected workflows built into Atomic, with detailed guides to implementation and design workflows plus six reusable control-flow patterns.
Recursive state machines and coding agent execution
Separates finite state machines, recursive state machines, and recursive language models from the acyclic execution graph Atomic materializes, validates, and bounds.
Customer case studies
Case Study: GRiDD Technologies
GRiDD Technologies built jira-ship on Atomic and ran twelve Jira tickets through it: 18 of 19 generated PRs merged with zero follow-up fixes and zero reverts, roughly 375 verification commands ran inside the workflows, and one $4.85 run caught a change that would have silently deactivated locations for paying accounts.
Maintained documentation
Atomic documentation
Complete hosted documentation for installation, configuration, agents, tools, extensions, and runtime behavior.
Workflow documentation
The source of truth for workflow discovery, authoring, primitives, contracts, run controls, composition, and built-ins.
Agent-readable documentation index
A maintained technical-documentation map for coding agents and other text-first clients.
Source, examples, and releases
Atomic on GitHub
The open-source repository, README, issue tracker, package source, and contribution history.
Repository workflow definitions
Substantial TypeScript workflows used by the Atomic project, including release operations with deterministic gates.
Workflow package source
The public authoring surface, runtime implementation, built-ins, and package-level API examples.
Atomic releases
Published versions and release notes. Check them when behavior may depend on the installed version.
Selected original field notes
These articles from the Mixture of Experts Blog are historical, opinionated accounts, not API documentation. They provide design context and examples without replacing the maintained Atomic pages above.
Atomic’s Workflow SDK: deterministically extending coding agents
Argues that a workflow layer should wrap a coding agent rather than replace it, and that repo-owned TypeScript makes the process inspectable. It describes Atomic as an SDK around a separate agent harness, which predates Atomic becoming its own CLI.
Workflows are the new interface for coding agents
Argues for developer-owned orchestration, explicit handoffs, and workflow visibility beyond a single chat session.
Building a Loop Engine
A practical account of separating the agent’s inner reasoning loop from a typed, observable, and resumable outer workflow for production engineering work.
Atomic’s Ralph loop
Records the design history of an iterative planning, implementation, and review loop. Treat implementation details as time-bound and verify them against current documentation and source.
Plausible code is not correct code
Explains why tests must cover semantic and performance claims, not only compilation and happy-path behavior.
Recursive State Machines for Language Models
Argues that the unit of recursion should be a tracked execution stage rather than a nested model call. For the current implementation boundaries, read the recursive state machines guide alongside it.
How this index is curated
This editorial index is an overview; it does not reproduce the full documentation. The complete technical documentation is the source of truth for workflow APIs and current runtime behavior. Use source for implementation evidence, releases for version changes, and field notes for historical reasoning.