Summary
18 published capabilities (owner.team: "callweave") have no capability-src/ source in this repo at all — only capabilities/<ns>/<id>/1.0.0/contract.json + an artifact digest/URL pointing at a compiled .wasm release asset. This matches this repo's existing "WASM/source artifacts are never committed directly" architecture, but it means these 18 have no measurable test coverage here, unlike every other published capability.
Affected capabilities: approval.decision-apply, artifact.revision-create, classification.outcome-resolve, classifier.release-gate-evaluate, cluster.curate, coverage.completeness-assess, embedding.cluster-organize, location.initialize, model.improve, operations.recover, period.finalize, retention.eligibility-classify, review.prepare, review.state-transition-decide, revision.successor-assign, summary.aggregate, uncertainty.score, visual.parameter-map.
Why not backfilled by writing new tests ourselves
Considered and rejected: writing clean-room Rust reimplementations from each contract's own schema/use_cases (fully specified enough to do this) would prove an equivalent implementation is covered, not that the actual published .wasm bytes are — this repo has been careful never to overclaim verification (see decision-log entry 61, which stopped trusting a self-reported "passed" instead of independently checking). A fabricated coverage badge on borrowed logic isn't worth more than an honest "not measured."
Definition of Done
Not blocking
Same diff-based treatment as every other existing-content check in scripts/ci/capability_validation.py — the new coverage gate only judges newly-added contract.json files going forward, never re-litigates these 18. This is a cross-team ask, paced by Callweave's own availability, not a blocker for other registry work.
Summary
18 published capabilities (
owner.team: "callweave") have nocapability-src/source in this repo at all — onlycapabilities/<ns>/<id>/1.0.0/contract.json+ an artifact digest/URL pointing at a compiled.wasmrelease asset. This matches this repo's existing "WASM/source artifacts are never committed directly" architecture, but it means these 18 have no measurable test coverage here, unlike every other published capability.Affected capabilities:
approval.decision-apply,artifact.revision-create,classification.outcome-resolve,classifier.release-gate-evaluate,cluster.curate,coverage.completeness-assess,embedding.cluster-organize,location.initialize,model.improve,operations.recover,period.finalize,retention.eligibility-classify,review.prepare,review.state-transition-decide,revision.successor-assign,summary.aggregate,uncertainty.score,visual.parameter-map.Why not backfilled by writing new tests ourselves
Considered and rejected: writing clean-room Rust reimplementations from each contract's own schema/use_cases (fully specified enough to do this) would prove an equivalent implementation is covered, not that the actual published
.wasmbytes are — this repo has been careful never to overclaim verification (see decision-log entry 61, which stopped trusting a self-reported "passed" instead of independently checking). A fabricated coverage badge on borrowed logic isn't worth more than an honest "not measured."Definition of Done
maintainers@callweave.local) contributes real Rust source intocapability-src/<capability-id-with-dots-replaced-by-dashes>/for each of the 18, achieving the same bar this repo now requires of new publishes (functions=100%, lines/regions≥95% viacargo llvm-cov)gather_catalog_data.py'sCURRENT_CRATE_FOR_IDupdated to map each of the 18 once source landsNot blocking
Same diff-based treatment as every other existing-content check in
scripts/ci/capability_validation.py— the new coverage gate only judges newly-added contract.json files going forward, never re-litigates these 18. This is a cross-team ask, paced by Callweave's own availability, not a blocker for other registry work.