wf-composer

To convert natural language to a reusable pipeline, this Skill parses informal workflow descriptions, maps steps to concrete executors, injects checkpoints, and persists the final DAG as a JSON template for reuse. It requires an informal description of your intended workflow steps and input variables to begin parsing intent.

To generate a checkpointed DAG template, provide a semantic workflow description and the Skill will parse intent, resolve executors into a node graph, and automatically inject checkpoint nodes at phase and artifact boundaries. You then confirm or edit the visualized pipeline before it saves the template.

Yes, you can add review gates by requesting optional user pauses during workflow composition. The Skill injects checkpoint nodes at meaningful phase boundaries, allowing you to pause for review and confirm the pipeline before persisting the reusable JSON template.

The best way to orchestrate agents and CLI commands is using executor resolution, which maps each workflow step to concrete skill, CLI, agent, or command executors. This creates a node graph that supports parallel groups and optional agent delegation within the DAG template.

No specific dependencies are required to save templates. The Skill uses YAML frontmatter-driven discovery and phase-driven file outputs to persist the final DAG as a JSON template directly to a local template library in the .workflow/templates directory.

You should avoid natural language to pipeline generation if your workflow requires highly rigid, non-editable execution paths without user confirmation. The Skill relies on a confirmation and editing loop to visualize, add, remove, or edit nodes before persisting, which may not suit fully unattended automated pipelines.