Workflows That Write Code
CodingAgentStep is an autonomous code-generation step type for workflows. Designed for sandboxed execution, multi-language, and fully auditable — it clones your repo, writes code, runs tests, and produces a pull request with a clear diff. Currently paused on all plans.
How It Works
CodingAgentStep in the workflow builder
Define a natural-language task, pick a model (Claude Sonnet 4.6 recommended), and configure sandbox settings. The agent clones your repo, writes code, runs tests, and creates a pull request — all inside a single workflow step.
- Natural-language task description
- Model selection per step
- Designed for isolated sandbox execution
- Automatic PR creation
Sandbox (by design)
Isolated execution, by design
Each execution is designed to run in an isolated container with no access to the host filesystem or network, configurable memory (up to 2 GB), CPU time limits and an execution timeout, with every file change captured. Completing this isolation is the work under way before the step is re-enabled.
- Isolated filesystem per execution
- No network access to host
- Configurable resource limits
- Full audit trail of changes
Real-World Use Cases
From test generation to framework migrations
Auto-generate unit tests for untested modules, create documentation from code, migrate between frameworks, generate boilerplate from specs. Each execution produces a PR with a clear diff.
- Test generation and coverage improvement
- Documentation sync with code changes
- Framework and dependency upgrades
- Boilerplate generation from specifications
How It Works
Three steps to autonomous code generation
Add CodingAgentStep
Add a coding agent step to any workflow. Describe the task in natural language — "write unit tests for the auth module" or "migrate the API client from v2 to v3."
Configure sandbox
Set the model, memory limits, timeout, and target repository. The agent clones the repo and begins execution in a sandbox designed to isolate each run.
Review output
The agent produces a pull request with code changes, test results, and execution traces. Review the diff, check the test output, and merge when ready.
Plans
Availability
CodingAgentStep is currently paused on all plans. The tiers below describe availability once it returns.
Starter (Free)
- CodingAgentStep not available
- Not included
Pro / Team
- CodingAgentStep in workflows
- Isolated sandbox execution (on return)
- Automatic PR generation
- 10 executions per day
Enterprise
- Unlimited executions
- Custom container images (on return)
- VPC agent support
- Everything in Pro / Team
Want to know when CodingAgentStep returns?
CodingAgentStep is paused on all plans while we move its execution into isolated, per-run sandboxes. Get in touch and we will let you know when it is available again.