gxwf draft-next-step

Deterministic next-step picker for the per-step authoring loop. Wraps nextDraftStep from @galaxy-tool-util/schema. Native (.ga) input is rejected — drafts are format2-only.

The harness loop reads as: while (next = gxwf draft-next-step <wf>).draft: invoke per-step skill on next.step.

Output

Default output is pretty JSON — the agent-loop wire format. Same input → byte-identical output. --output-format markdown renders a human-glance checklist instead.

JSON fields:

• draft (bool) — true while a drafty step remains, false at the terminal cases below.
• step (string array) — the picked step’s path: a single id at top level (["merge assembled transcripts"]), or the full path through containing subworkflow steps when the pick is nested. The leaf (last element) is the step to work on.
• work (string array) — the chosen step’s remaining TODO[...] sentinels followed by its _plan_* fields, in checklist order.

At the terminal cases the object is just { "draft": false } — no step or work keys.

Two terminal cases both surface as draft: false:

• The workflow’s class is already GalaxyWorkflow (concretization complete, class promoted).
• The class is still GalaxyWorkflowDraft but every step is fully resolved (no TODO_*, no _plan_*). The harness’s next move is draft-extract (which calls promoteFullyConcreteDrafts to flip the class) followed by terminal validate.

Examples

gxwf draft-next-step workflow.gxwf.yml
gxwf draft-next-step workflow.gxwf.yml --output-format markdown

Harness loop sketch:

while [ "$(gxwf draft-next-step workflow.gxwf.yml | jq -r .draft)" = "true" ]; do
  step=$(gxwf draft-next-step workflow.gxwf.yml | jq -r '.step | last')
  # invoke per-step skill on step, which mutates workflow.gxwf.yml in place
done

Gotchas

• Pick order is topological (a step waits until its in: sources are concretized) with alphabetical tiebreak on step id at the same level. Same input across runs returns the same pick — safe to call repeatedly without state.
• Descends into draft subworkflows; the returned step array is the full path through containing subworkflow steps (leaf last).
• A draft: false result does not mean the workflow has passed validate yet. It means the per-step authoring loop is done; the harness still owns the promote + terminal-validate pass.
• This command does not mutate the workflow file. It’s a pure observer.
• Insertion order of step dicts does not affect output — semantically equal workflows produce identical picks.