workflow-authoringlisted

# Workflow Authoring Treat the marketplace as a casebook of reusable orchestration designs. The built-in Workflow tool already supplies the execution model; this skill adds authorship judgment: deciding whether a workflow is warranted, selecting compatible patterns, adapting them without copying accidental details, and making the resulting script honest about cost, uncertainty, failure, and coverage. Prefer a small script with one defensible idea over a miniature workflow framework. A workflow earns its overhead when several agents can contribute genuinely different evidence or when an explicit trust structure—independent verification, a completeness critic, falsification, or a bounded repair loop—materially improves the result. Recommend an ordinary prompt or one agent when the work is short, tightly sequential, or cannot be partitioned without duplicating context. This skill also provides an optional user-facing design process. Its value is not that a user must out-author Claude; it makes Claude’s orchestration choices inspectable and revisable before the workflow is trusted. When the user invokes the skill or `/wf-new`, expose the major decisions, teach the few composition patterns that matter to this design, and show the static diagram rather than silently producing a script. ## Reference routing Read only what the task needs, but read the selected reference completely. | Reference | Use it for | |---|---| | `references/catalog-patterns.md` | Start here when selecti