Code health receipts
September 7, 2026 · View on GitHub
Kody is a source-available codebase written almost entirely by AI agents. Instead of asking readers to take code quality on faith, this page records the measured numbers and the enforcement that keeps them from regressing. All figures below were measured directly against the repository (2026-08-30).
Measured baseline
| Signal | Value |
|---|---|
as any casts in packages/ | 1 |
TODO comments in packages/ | 0 |
@ts-expect-error suppressions in packages/ | 1 |
| Duplicated lines (jscpd, min 70 tokens, non-test source) | 1.96% |
| Runtime dependencies of the main worker package | 28 |
Decision records in docs/contributing/decisions/ | 45 |
Checks in the npm run validate gate | 23 |
Enforcement, not promises
npm run validateis the single authoritative gate: format, lint, typecheck, node/workers/e2e/mcp test suites, build checks for every worker, migrations, deploy guardrails, temporal-docs and decision-record checks, the file-size ratchet, and knip dead-code analysis.tools/file-size-ratchet.jsonenforces budgets of 800 lines for client routes and 2,000 lines for node test files. Files may only shrink out of the grandfathered list; new oversized files fail CI.knipfails the gate on unused files, exports, and types.tools/check-decorative-banners.tsrejects decorative comment banners.
Oversized-file cleanup receipts
Every grandfathered file over budget was split along behavior boundaries with no runtime or assertion changes, and its ratchet exception removed in the same PR. The ratchet covers each of those files.
Client routes (budget: 800 lines)
| Route | Before | After | PR |
|---|---|---|---|
| connect-oauth.tsx | 2,192 | 795 | #1852 |
| account-integrations.tsx | 2,167 | 581 | #1851 |
| onboarding.tsx | 1,628 | 713 | #1853 |
| account-secrets.tsx | 1,610 | 777 | #1854 |
| community-detail.tsx | 1,490 | 800 | #1865 |
| onboarding-mcp-client-tabs.tsx | 1,477 | 556 | #1869 |
| account.tsx | 1,440 | 791 | #1855 |
| admin-users.tsx | 1,435 | 784 | #1866 |
| login.tsx | 1,335 | 773 | #1870 |
| admin-insights.tsx | 1,289 | 134 | #1871 |
| account-mcp-servers.tsx | 1,251 | 598 | #1858 |
| account-jobs.tsx | 1,146 | 554 | #1872 |
| admin-codemods.tsx | 1,100 | 789 | #1875 |
| admin-platform-integrations.tsx | 1,098 | 565 | #1877 |
| account-activity.tsx | 954 | 516 | #1878 |
| account-email.tsx | 891 | 487 | #1861 |
| record-table.tsx (shared table module) | 837 | 586 | #1863 |
| home.tsx | 808 | 562 | #1857 |
Extraction pattern: routes keep loader consumption, mutations, and lifecycle orchestration; extracted modules are render-only sections/forms/detail panels plus shared URL/filter helpers.
Node test suites (budget: 2,000 lines)
| Suite | Before | After (entrypoints) | PR |
|---|---|---|---|
| jobs/service.node.test.ts | 5,596 | 5 workflow suites, 723–1,286 each | #1864 |
| package-runtime/module-graph.node.test.ts | 3,093 | 990 + 1,161 + 892 | #1879 |
| mcp/run-kody-registry.node.test.ts | 2,889 | 1,039 + 1,305 | #1880 |
| package-invocations/service.node.test.ts | 2,857 | 1,212 + 820 | #1881 |
| repo/repo-session-do.node.test.ts | 2,764 | 1,288 + 1,193 | #1882 |
| package-runtime/package-workflows.node.test.ts | 2,534 | 999 + 1,239 | #1883 |
| app/account-deletion.node.test.ts | 2,234 | 1,110 + 615 | #1884 |
| account/export.node.test.ts | 2,147 | 771 + 1,234 | #1885 |
Every split preserved every test and assertion (verified by matching test counts
before and after), moved reusable fixtures to
packages/worker/src/test-support/, and kept vi.mock wiring file-local
because Vitest hoists mocks per test file. No production code changed.
What was intentionally not done
- No Durable Object rewrites: large DOs are frozen by the ratchet and shrink opportunistically.
- No test behavior changes: splits reorganize, never weaken, coverage.
- No suppression sweeps: the near-zero
as any/TODOcounts are the actual state of the tree, not a lint-ignore artifact.