Context
#239 vendored the full FileMaker XML Snippet Claude Skill (v1.12) and two Clockwork Inspector extracts under docs/filemaker-xml-canonical/ — about 192K of markdown/JS (168K skill + 24K inspector), on top of SharpFM's own tooling and review notes in the same directory.
None of it is read at build or test time. refresh-canonical.ps1 re-syncs the vendored copies from upstream, and docs/filemaker-xml-canonical/extract-fixtures.cs regenerates the 253 checked-in XML fixtures under tests/SharpFM.Tests/CanonicalSkill/fixtures/ from the skill markdown — those fixtures (not the vendored source) are what the round-trip test suite actually reads. Both scripts are dev-time only, not wired into CI.
Question
Is vendoring the full upstream source worth the repo weight, versus a lighter alternative — e.g. only keeping the generated fixtures plus the refresh script fetching upstream content on demand, or trimming the vendored copy to just the files that feed fixture generation (dropping pure narrative like skill/references/worked-example.md and skill/README.upstream.md)?
Ask
Track how often refresh-canonical.ps1 / the vendored reference actually gets consulted in practice (e.g. next FileMaker release bump) before deciding. Revisit this issue once there's real usage signal one way or the other.
Context
#239 vendored the full FileMaker XML Snippet Claude Skill (v1.12) and two Clockwork Inspector extracts under
docs/filemaker-xml-canonical/— about 192K of markdown/JS (168K skill + 24K inspector), on top of SharpFM's own tooling and review notes in the same directory.None of it is read at build or test time.
refresh-canonical.ps1re-syncs the vendored copies from upstream, anddocs/filemaker-xml-canonical/extract-fixtures.csregenerates the 253 checked-in XML fixtures undertests/SharpFM.Tests/CanonicalSkill/fixtures/from the skill markdown — those fixtures (not the vendored source) are what the round-trip test suite actually reads. Both scripts are dev-time only, not wired into CI.Question
Is vendoring the full upstream source worth the repo weight, versus a lighter alternative — e.g. only keeping the generated fixtures plus the refresh script fetching upstream content on demand, or trimming the vendored copy to just the files that feed fixture generation (dropping pure narrative like
skill/references/worked-example.mdandskill/README.upstream.md)?Ask
Track how often
refresh-canonical.ps1/ the vendored reference actually gets consulted in practice (e.g. next FileMaker release bump) before deciding. Revisit this issue once there's real usage signal one way or the other.