Skip to content

Stop run-workflow-test guessing a Galaxy branch#363

Merged
jmchilton merged 1 commit into
mainfrom
close-361-no-guessed-galaxy-branch
Jul 20, 2026
Merged

Stop run-workflow-test guessing a Galaxy branch#363
jmchilton merged 1 commit into
mainfrom
close-361-no-guessed-galaxy-branch

Conversation

@jmchilton

Copy link
Copy Markdown
Member

Closes #361.

Problem

The sarek run authored a workflow with sample_sheets, then tested it against Galaxy release_25.0. Tracing it back (thanks @mvdbeek), the branch was introduced by the phase-12 agent when it generated run-planemo.sh — not by a planemo default, a Foundry source note, or an earlier artifact. The run-workflow-test Mold left the Galaxy version entirely unspecified, and the agent filled the vacuum with a guess.

Fix

Close the vacuum, minimally.

  • content/molds/run-workflow-test/index.md — new step 3: leave --galaxy_branch off unless the user or harness supplied one, never guess a release_*, and record a supplied branch in the output so a later failure can be read against it.
  • content/research/planemo-workflow-test-architecture.md — one factual line under managed-mode behavior: --galaxy_branch defaults to master, so omitting it tests newest; naming an older release narrows accepted workflow syntax and the rejection surfaces as an invocation-time error rather than a version mismatch. Five Molds load this note on demand, so the fact also lands where a debug pass looks.

Deliberately not done: no galaxy_branch frontmatter field, no version-resolution logic, no pinned release anywhere. A pin would go stale and reproduce this bug on a slower clock.

Casts

Recast the five bundles carrying the note. freeform-summary-to-galaxy-test-plan also picks up a pre-existing stale schema reconciliation (the galaxy source-family enum, already in main's package — that cast just hadn't been re-run).

Checks

npm run validate — 0 errors. make check-assemble-pipelines — clean.

npm run test has 2 failures in tests/cast-mold.test.ts (summarize-nextflow schema drift). These are pre-existing on main, confirmed by stashing and re-running; untouched here.

🤖 Generated with Claude Code

Mold left the Galaxy version unspecified; a run filled the vacuum with
--galaxy_branch release_25.0, which rejected the workflow at invocation.

Add explicit no-pin rule to the Mold and record planemo's master default
in the architecture note. No new frontmatter field, no pinned release -
a pin would go stale and reproduce this on a slower clock.

Recast the five casts bundling the note. freeform-summary-to-galaxy-test-plan
also picks up a pre-existing stale schema reconciliation.

Closes #361

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@jmchilton
jmchilton merged commit 2180d6a into main Jul 20, 2026
1 check failed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Ran old Galaxy in nextflow sarek test

1 participant