[dinwwwh]

🤖 Generated with Claude Code

https://claude.ai/code/session_01VVaw7RYWCsn8myBE4Rncuq

Test-only. I applied each surviving Stryker mutant to a copy of the source and compared its output with main's on 50k fuzz docs. Six mutants change output, and no test caught them:

• isSameBranch's name === 'enum' && isOnlyNull(a[name]) (3 mutants). These take { type: [string, null], enum: ['a'] } for the null branch beside it, so the pair merges into { type: string, enum: ['a'], nullable: true }. That output loses null, because the enum holds no null. A new assertion pins the anyOf of both.
• '$ref' in NOT_NULLABLE_KEYWORDS → "". This mutant lets a target written as { $ref: Base, type: [object, null] } count as admitting null, so a pair to it collapses to the bare $ref, which loses null when Base rejects it. A new assertion pins that the pair stays.
• out.nullable !== true in finishSchema (2 mutants). These mark { type: [array, null], nullable: true } as loose, so a oneOf with a $ref to it becomes an anyOf. A new assertion pins the oneOf.

Each assertion passes on main and fails under its mutants. I checked the rest by hand. The other 4 survivors that change output are equivalent: 2 only reorder keys (patternProperties/prefixItems guards), 1 was a mis-parenthesized copy of a mutant, and 1 is the known binary-format baseline survivor.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

✅ 29 untouched benchmarks
⏩ 2 skipped benchmarks¹

---

_{Comparing claude/project-thread-rixcs1 (abcbc8f) with main (e581541)}

Footnotes

1. 2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[pullfrog]

✅ No new issues found.

Reviewed changes — a test-only PR adding three assertions to packages/downgrader/tests/v3.1-to-v3.0.test.ts that pin converter behavior previously unguarded against surviving Stryker mutants.

• Enum branch beside null — pins that anyOf: [{ type: ['string','null'], enum: ['a'] }, { type: 'null' }] stays a pair, guarding the isOnlyNull exception in isSameBranch so an enum that rejects null is not treated as the null branch.
• $ref beside a nullable type list — pins that a nullable pair targeting { $ref: Base, type: ['object','null'] } stays a pair, guarding '$ref' in NOT_NULLABLE_KEYWORDS so the target's $ref (which rejects null) is honoured.
• Nullable type list's looseness — pins that oneOf: [{ $ref: T }, { type: 'integer' }] with T = { type: ['array','null'], nullable: true } keeps its oneOf, guarding out.nullable !== true in finishSchema so T is not marked loose.

I re-ran the test file on the unmodified source (137 passing) and applied each of the three mutations in place: every one makes exactly the intended assertion fail, so none of the new tests is vacuous. uncheck on the changed file also passes oxlint, oxfmt, and tsc. The body's survivor arithmetic is consistent, and the assertions match the existing comment density and idiom of the surrounding tests.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}