chore(lint): adopt oxlint 1.67 canonical rule additions (selective, library posture)

[Goosterhof]

Summary

Selective adoption of the war-room canonical template's oxlint 1.55→1.67 rule additions (template commit 321da5f), respecting fs-packages' deliberate library Correctness-only posture (per CLAUDE.md §Lint Rules: "library posture is Correctness-only, opt-in per-rule for anything else").

This is not a wholesale port — Vue/jsx-a11y/type-aware bulk rules are deliberately out of scope and the skip rationale is documented inline in .oxlintrc.json.

Standalone — no longer stacked. This PR was originally drafted stacked on #98 (Supply Corps sweep, which carried the oxlint 1.67 binary). #98 was closed without merging (2026-05-29); oxlint 1.67 is already on main regardless (package.json pins "oxlint": "^1.67"), so this PR stands alone, is MERGEABLE against main directly, and needs nothing merged first.

Gating note: the CI lint step runs bare npm run lint (oxlint .) — not --deny-warnings. The three warn-severity rules below (no-unnecessary-type-parameters, no-useless-iterator-to-array, import-style) are therefore advisory in CI: they surface in output but exit 0. They are documentation-of-intent, not enforcement, unless/until the CI gate adds --deny-warnings. The four error-severity rules gate normally.

Rules adopted (7)

Rule	Severity	Rationale
no-implied-eval	error	Core ESLint security — fires on setTimeout("...") / setInterval("...") / new Function("...") callsites. No posture friction.
prefer-regex-literals	error	Core ESLint correctness — new RegExp("foo") → /foo/.
getter-return	error	Graduated from nursery → correctness (1.61). Catches getters with missing return.
typescript/no-unnecessary-type-conversion	error	Library code benefits from minimal String(x)/Number(x)/Boolean(x) noise on already-typed values.
typescript/no-unnecessary-type-parameters	warn (advisory in CI)	Type parameter used only once. API-surface smell.
unicorn/no-useless-iterator-to-array	warn (advisory in CI)	Pure performance — flags [...someArray] when result is iterated again.
unicorn/import-style	warn (advisory in CI)	Pushes the canonical import shape per node builtin (e.g. import path from 'node:path', not import {join}). Complements prefer-node-protocol.

Deliberately NOT adopted (documented inline in .oxlintrc.json)

Skipped	Why
vue/*	No SFCs in fs-packages source; published Vue-aware packages expose factories/composables only.
jsx-a11y/*	No JSX in this monorepo.
typescript/no-unsafe-* family	Library API surface intentionally exposes unknown/any at adapter boundaries (fs-adapter-store generic constraints).
unicorn/prefer-import-meta-properties	Dual ESM+CJS build target via tsdown — import.meta in CJS bundles cascades to build errors.
no-useless-assignment	Library code intentionally pre-declares values for type inference.
vitest/valid-expect-in-promise, vitest/prefer-strict-boolean-matchers	Test gates already enforce strict matcher discipline via Stryker mutation testing at 90% threshold.
Vue 2→3 deprecation guards, Composition-API safety, Options-API correctness	No Options API or template-API features used.
complexity, max-params, max-depth, max-nested-callbacks, max-lines-per-function	Stryker mutation + 100% per-package coverage gates govern complexity at the test-discipline layer.

Sweep collateral (in-scope, ≤3 line fix per order §C)

scripts/lint-pkg.mjs line 25 — unicorn/import-style fires on import {join} from 'node:path' with diagnostic "Use default import for module node:path". Rule is bidirectional per-builtin (some builtins want named, some want default); for node:path it wants default. Switched to import path from 'node:path' with path.join(...) at the 4 call sites in this file. Behaviour unchanged. 5-line diff.

Note: the order's §A rationale for unicorn/import-style describes the rule as "pushes named imports for node builtins" — directionally wrong for node:path. Fix matches what the rule actually fires on. Flagged for canonical template maintainer in the execution report.

Verification (8-gate matrix, run locally)

Gate	Status
1. npm audit (--high)	PASS — 2 moderate transitive (qs/typed-rest-client) below --audit-level=high; pre-existing baseline carried from Supply Corps sweep
2. npm run format:check	PASS — 145 files, no diffs
3. npm run lint	PASS — zero output; --deny-warnings also exits 0 today (no current warn-rule violations), but --deny-warnings is NOT the CI gate
4. npm run build	PASS — all 11 packages built clean
5. npm run typecheck	PASS — all 11 packages clean
6. npm run lint:pkg	FAIL — PRE-EXISTING BASELINE (22 fails: sideEffects advisory × 11 [queue #70] + attw 0.18.2 "Cannot read properties of undefined (reading 'filename')" × 11 [PR #98 commentary]). Reproduced identically on parent branch — not introduced here.
7. npm test	PASS — 528/528 tests across 22 test files
8. npm run test:coverage	PASS — per-package 100% thresholds enforced and met

Mutation testing not run — config-only change does not move the mutation surface; per-package Stryker config unchanged. 7/8 gates green; Gate 6 carries the documented pre-existing baseline that PRs #87, #88 and the 2026-05-15 wave already flagged.

Residual baseline carried forward (NOT introduced)

• lint:pkg Gate 6 — 22 fails identical to parent branch chore/supply-corps-sweep-2026-05-26. Tracked: enforcement queue fs-dialog: forward host <dialog> ARIA attributes via dialog.open() options (#67) #70 (sideEffects manifest compliance — but note PR chore: declare sideEffects: false across all 11 packages — closes queue #70 #88 already adopted "sideEffects": false, the suggestion fires because the wrapper packs from source-tree, not from dist/ — separate concern) + attw 0.18.2 known regression (PR chore: Supply Corps sweep — npm refresh + audit fix (3 advisories cleared) #98 commentary).
• npm audit — 2 moderate (qs, typed-rest-client) below --audit-level=high. Carried from Supply Corps sweep PR chore: Supply Corps sweep — npm refresh + audit fix (3 advisories cleared) #98 npm audit fix run.

Commit shape

Single commit: chore(lint): adopt oxlint 1.67 canonical rule additions selectively.
2 files changed, 26 insertions, 6 deletions.

Execution report

reports/fs-packages/execution/2026-05-26-armorer-oxlint-1.67-adoption.md in the war-room repo. Includes per-skip rationale, full 8-gate matrix, and Notes-for-the-General on the rule-direction wording issue.

Test plan

• npm run lint (oxlint w/ new rules) exits 0
• npm run lint -- --deny-warnings also exits 0 (local check only — not the CI gate)
• npm run build clean
• npm run typecheck clean
• npm test 528/528
• npm run test:coverage clean
• lint:pkg failure matches parent branch (confirmed pre-existing, not introduced)

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying fs-packages with    Cloudflare Pages

Latest commit:	71ab0b4
Status:	✅  Deploy successful!
Preview URL:	https://9c22a655.fs-packages.pages.dev
Branch Preview URL:	https://chore-oxlint-1-67-canonical.fs-packages.pages.dev

View logs

[jasperboerhof]

PR Reviewer · 9/10 · PASS

• PR: chore(lint): adopt oxlint 1.67 canonical rule additions (selective, library posture) #99 · fs-packages
• AC anchor: none
• Reviewers run: Acceptance, Simplicity, Surface, Silent failures skipped — no triggers, Efficiency skipped — no triggers
• Acceptance: SKIP
• Simplicity: PASS · 9/10
• Surface: PASS · 9/10

Findings

• MINOR · .oxlintrc.json:24 — no-useless-assignment is named in two separate skip-rationale comments (line 21 and line 24) with mildly inconsistent justifications — line 21 says "intentionally pre-declares values for type inference", line 24 bundles it with complexity/max-* limits saying "Stryker + coverage gates govern complexity". Drop the duplicate mention on line 24 (keep it grouped with the complexity/max-* limits, or keep line 21 — but not both) so the skip-list reads cleanly.

Action

merge-ready

[Goosterhof]

No blockers. Two concerns and two nits on a clean, well-documented config chore. The diff is exactly what the body advertises — 7 selectively-adopted oxlint rules with inline skip rationale, plus a 5-line node:path import-shape fix in scripts/lint-pkg.mjs. The library Correctness-only posture is respected and the skip table is honest about why the bulk rules stay out.

Concerns

The "Stacked on #98" narrative is stale and now actively misleading. The body says this PR is "Stacked on #98 (Supply Corps sweep — carries the oxlint 1.67 binary the new rules depend on)" and that "Once #98 merges this PR rebases onto fresh main." But #98 is CLOSED and was never merged (mergedAt: null, mergeCommit: null, closed 2026-05-29). The dependency the stacking narrative hangs on — oxlint 1.67 — is already on main regardless (package.json on main pins "oxlint": "^1.67"), and this PR reports MERGEABLE against main directly. So the stacking framing is no longer load-bearing: there's nothing to merge first and nothing to rebase onto. A reviewer reading the body will go looking for an open #98 and find a closed one. Rewrite the Summary to state the actual situation — #98 closed, oxlint 1.67 already on main, this PR stands alone — so the next reviewer doesn't have to reconstruct the merge history to trust the diff.

unicorn/import-style set to warn doesn't gate by default. Three of the adopted rules (no-unnecessary-type-parameters, no-useless-iterator-to-array, import-style) are warn, and the default lint script is oxlint . without --deny-warnings. So in normal CI these three are advisory only — they surface in output but exit 0. The body's verification matrix notes --deny-warnings also exits 0 today, which is the right thing to have checked, but unless --deny-warnings is the actual CI gate (not just a local spot-check), a future violation of these three lands green. Either confirm the CI lint step runs --deny-warnings (in which case warn and error are equivalent for gating and the severity split is purely cosmetic), or accept that these three are documentation-of-intent rather than enforcement. Worth one sentence in the body stating which it is.

Nits

The skip-rationale comment block lives inside the "rules": {} object as trailing // lines after the last real entry. It parses fine (oxlint reads JSONC and npm run lint passing proves it), but a trailing comment block with no following key is the kind of thing a future editor deletes by accident when adding the next rule. Consider a single leading block comment instead of a trailing one.

The node:path fix switches {join} to a default path import to satisfy import-style's per-builtin direction. Fine, behaviour-identical. The body already flags that the canonical template's §A rationale describes the rule as pushing named imports — directionally wrong for node:path — and routes that to the template maintainer. Good catch; nothing to change here.

Verdict

This is mergeable on substance. The only thing I'd want addressed before merge is the stale #98 narrative in the body — it's not a code defect but it's a correctness defect in the PR's own provenance story, and provenance is what a config-chore review trades on. Fix the body, confirm the --deny-warnings gating question, and ship.

[jasperboerhof]

Approved — reviewed, no blockers.