oxfmt and oxlint gate the v2 TypeScript

Context

The v2 gate checked Rust with cargo fmt and cargo clippy -D warnings, but TypeScript only with tsc. The house conventions (function declarations for standalone functions, arrow callbacks, no silent catch, Props for a component's only local type) existed only as written guidance, and formatting had drifted between an 80- and a 100-column style.

The anti-slop Oxlint plugin offers stricter type-evidence rules. Applied in full to v2 it reports over 1,200 findings, most of them legitimate boundary code: typeof guards and unknown parameters where IPC, storage, and sync payloads are decoded, and conditional spreads that omit optional fields.

Decision

Format and lint every v2 TypeScript and JavaScript source from the repository root with oxfmt (.oxfmtrc.json, 100 columns) and oxlint (.oxlintrc.json). Both run in tools/scripts/build.sh, so ./bin/check and CI fail on any finding. v1/ keeps its own configuration.

The rule set is:

• oxlint's correctness category plus explicit rules for the house conventions: func-style (declarations; a typed variable may hold an arrow when it implements a function-type contract), prefer-arrow-callback (named function expressions stay allowed so memo and forwardRef components keep their display names), no-explicit-any, type-only imports, type over interface, and react/no-unstable-nested-components.
• A vendored subset of anti-slop in tools/oxlint/anti-slop/: chained type assertions, widen-then-assert, Reflect.get/Reflect.apply, module mocking, object parameters, and aliases that resolve to unknown.
• A local skriuw plugin in tools/oxlint/skriuw/: no-silent-catch (a catch must contain a statement; comments do not count, noop() marks a deliberate swallow) and local-type-name (a .tsx module's single, non-exported type is named Props; other modules keep descriptive names because Props means component props).

Test code, harnesses, and e2e drivers may use any and chained assertions for test doubles.

Deferred

These rules are off, not rejected. Each needs a behavior-reviewed migration before it can gate:

• React Compiler rules (react/refs, set-state-in-effect, purity, immutability, and related) and exhaustive-deps: roughly 190 findings in code that reads refs deliberately under the performance contract.
• jsx-a11y: roughly 150 findings, mostly role and focus semantics that change interaction behavior when fixed.
• anti-slop boundary rules (no-runtime-typeof, no-unknown-parameters, no-unknown-returns, no-unsafe-dictionary-type, require-safety-comment-for-type-assertion): these conflict with decoding at trust boundaries until those boundaries have schema parsers.

Consequences

• Style review becomes mechanical; the reformat commit is listed in .git-blame-ignore-revs.
• Suppressions need an inline reason and are visible in review.
• Upgrading oxlint requires upgrading @oxlint/plugins to the same exact version.