[devin-ai-integration]

What and why

internal/workspace/project.Bundle archived every non-hidden regular file under the directory, so sysml -bundle on the Open-MBEE/OpenSysML-Extensions-Library tree gave 36 members (README, CONTRIBUTING, scripts/**, validation/, engine-contract.json, even scripts/__pycache__/*.pyc) where the project's own scripts/build-kpar.py gives 19, although the two manifests were byte-identical.

The rule now: a directory with a .meta.json is archived as the two manifests plus exactly the files the metadata records — every checksum key and every index target, at their recorded paths. A recorded file missing from the tree is an error naming it (.meta.json records LICENSE, which is not in the directory); a recorded path that escapes the directory is refused (.meta.json records "../README.md": …); a recorded entry that is not a regular file is refused. The kept .meta.json is still verified on the read-back, so a stale checksum refuses the bundle as before. A directory without a .meta.json is archived as KerML 1.1 §10.3 describes a project archive — "a model interchange file for each of the root namespaces in the project" and the two manifests — so its members are .project.json and the model files (.sysml, .kerml) outside hidden entries, library directories and nested projects, with the .meta.json generated from them; a README, __pycache__, editor backups (X.sysml~), .git/ and build scripts never travel. A project that wants its LICENSE or anything else in the archive records it in a .meta.json, as the Extensions Library does.

Consequences stated plainly: an archive another tool wrote round-trips to the files its .meta.json records. sysand's mbse4u-sysmod_5.1.4.kpar carried a README and a license its metadata does not record, so unbundling and re-bundling it gives 5 members, not 7; the round-trip test now says so. The Bundle doc comment and docs/guide/03-command-line.md document the rule.

Acceptance: sysml -bundle against build-kpar.py and the published release

Clone of OpenSysML-Extensions-Library at b36b361f85af4764de8e2817cbae62fb9e74ffdb with scripts/__pycache__/x.pyc added:

$ sysml -bundle /tmp/accept -o /tmp/accept-ours.kpar
wrote /tmp/accept-ours.kpar (OpenSysML Extensions Library 1.0.0, 19 files)
$ python3 /tmp/accept/scripts/build-kpar.py --output /tmp/accept-theirs.kpar
Wrote /tmp/accept-theirs.kpar (OpenSysML Extensions Library 1.0.0, 19 entries)
$ curl -sfLO https://github.com/Open-MBEE/OpenSysML-Extensions-Library/releases/download/v1.0.0/opensysml-extensions-library-1.0.0.kpar

unzip -l of all three, sizes and names (the develop binary gives 36 members on the same tree):

3691 .meta.json
1343 .project.json
11357 LICENSE
3699 libraries/AnalysisRecords.sysml
4909 libraries/DiagramLayout.sysml
15931 libraries/DocumentQueries.sysml
680 libraries/IdentityMetadata.sysml
21622 libraries/MOSA.sysml
2765 libraries/MigrationMetadata.sysml
20884 libraries/OOSEM.sysml
1911 libraries/OpenSysMLMathFunctions.kerml
454 libraries/OpenSysMLRenderings.sysml
1972 libraries/RandomFunctions.kerml
3613 libraries/Simulation.sysml
1681 libraries/StateActivity.kerml
2572 libraries/StateMachines.sysml
2458 libraries/StateSpaceIntegration.sysml
1556 libraries/Stochastic.sysml
14660 libraries/SysMLValidation.sysml

diff of the three listings: empty. cmp of each of the 19 members extracted from ours against build-kpar.py's and against the published opensysml-extensions-library-1.0.0.kpar: every member identical. The published archive is byte-identical to build-kpar.py's (SHA-256 1eb10ec8…); ours differs from both only in the ZIP container headers (Go's writer records a Unix mode and extended-timestamp field where Python's ZipInfo(external_attr=0, create_system=0) does not), not in any member.

Specification basis

KerML 1.1 Beta 2 §10.3 (Model Interchange Projects): "The archive shall contain a model interchange file for each of the root namespaces in the project … In addition, the archive shall contain, at its top level, exactly one file named .project.json and exactly one file named .meta.json." No row in docs/project/spec-compliance.md moves.

How it was verified

• go test ./internal/workspace/project — TestBundleArchivesTheFilesTheMetadataRecords (recorded subset, missing recorded file named, ../ path refused), the metadata-generation test now expecting the model files alone (with an editor backup in the tree), TestBundleLeavesNestedProjectsOut, TestBundleLeavesTheLibraryDirectoryOut.
• go test ./tests/kpar — TestBundleIsDeterministicAndUnbundleReproducesTheRecordedFiles (renamed; compareRecorded compares the recorded set, the sysmod round trip expects 5 members) and the new tests/kpar/bundle_test.go: a repository-shaped tree built at test time (two units, LICENSE, README, CONTRIBUTING, scripts/build.py with __pycache__, Widgets.sysml~, .git/HEAD, .github/) whose .meta.json records three files bundles to exactly those plus the manifests, byte for byte; LICENSE removed → cannot bundle …: .meta.json records LICENSE, which is not in the directory; .meta.json removed → the project file and the two model files, index generated.
• The acceptance run above.
• Full gate on this branch: go build ./... ok, go vet ./... ok, gofmt -l . empty, make lint (staticcheck + gosec, both modules) ✓, go test ./... with OPENSYSML_REQUIRE_TRAINING_CORPUS=1 OPENSYSML_REQUIRE_PILOT_CORPORA=1 OPENSYSML_REQUIRE_SPEC_EXAMPLES=1 after the three download scripts — all packages ok, python3 scripts/changelog.py check ok, python3 scripts/check-doc-links.py → 0 broken links, python -m mkdocs build --strict ok, make extension-libraries-check → in sync.
• Dry-merges: clean against origin/develop, against chore(extensions): re-pin to the project-interchange upstream and prove the vendored copy is the published project #1563 (the sibling), and textually clean against chore(pilot): pin the OMG pilot implementation to release 2026-09 (0.63.0) #1529's head for every file this PR touches — the bundle section of tests/kpar/kpar_test.go does not overlap chore(pilot): pin the OMG pilot implementation to release 2026-09 (0.63.0) #1529's OMG tier-map edits; chore(pilot): pin the OMG pilot implementation to release 2026-09 (0.63.0) #1529's own conflicts with develop are in docs/project/* files this PR does not change.
• Devin Review: one finding (a recorded path through a directory link leading outside the project was archived), fixed by confining every recorded member with confinedTo as Unbundle does, thread resolved; the current head has no open finding of any severity, flags included.

Checklist

• make test and make lint pass locally
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/1e9164f57f814014a9f0aba35666e2a3
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/1e9164f57f814014a9f0aba35666e2a3?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring