Home pipeline · interview → galaxy

UPDATE-INTERVIEW → GALAXY

Interview-driven, edit-in-place modification of an existing Galaxy gxformat2 workflow via a reviewable change-set, reusing the per-step draft loop.

Phases

/ subway map 9 phases

Run this pipeline

/ stop-gap harness

Run this pipeline today via its assembled harness skill pipeline-update-interview-to-galaxy — install once (run guide), then in a Claude Code session:

/foundry-skills:pipeline-update-interview-to-galaxy

Each run gets its own ./<run-slug>/ working directory so artifacts stay namespaced and a second run won't clobber the first — the harness sets this up first (phase detail).

phases
9
auto casts
5
manual
4
loops
1

UPDATE-INTERVIEW → GALAXY

Interview-driven, edit-in-place modification of an existing Galaxy gxformat2 workflow. Every other Foundry pipeline is greenfield — it generates a workflow from a non-Galaxy source. This one consumes a Galaxy workflow already and changes it: swap a tool, bump a version, add a step, retune a parameter, expose an output, rewire a connection. The workflow is both the input and the output.

Approach: diff-and-patch

The existing workflow enters the pipeline already concrete. Phase 1 reads it into a structured summary-galaxy-workflow that anchors the interview. Phase 2 turns the interview into a reviewable, step-anchored change-set (the human approval gate). Phase 3 applies the change-set to the concrete workflow — direct edits inline, tool-introducing edits injected as drafty steps — and emits a galaxy-workflow-draft.

An edit is a drafty region. From phase 3 on, the pipeline reuses the greenfield machinery unchanged: advance-galaxy-draft-step drains any injected drafty steps and extracts the concrete workflow, then the test/validate/run/debug tail runs. Untouched steps ride through at Resolved tier, byte-stable.

This mirrors the greenfield altitude structure: the change-set plays the reviewable-brief role the interface/data-flow briefs play; apply-galaxy-workflow-changeset plays the topology-settling role the *-to-galaxy-template Molds play (both emit a draft).

Why edit-in-place, not regenerate

Regenerating discards the existing structure, provenance, and tests. Diffing preserves them: untouched regions are unchanged, so the workflow’s shipped tests are a meaningful regression baseline — the modified workflow must keep passing them except where an edit intentionally changes an output. That baseline is the distinctive advantage of this pipeline over “regenerate the whole thing.”

Relationship to other work

Phase 1 closes a known inventory gap (compare-against-iwc-exemplar previously lacked a structured view of the exemplar it diffs against, and the Foundry had no Galaxy-as-source summarizer). summarize-galaxy-workflow serves both.