chore: housekeeping & make eslint happy

[SukkaW]

Summary

Housekeeping phase 1: sweep existing ESLint issues.

Verification

• 686 lint errors -> 0
• 454 lint warnings -> 415
• all existing passed tests still pass after the changes
• two callback refs are introduced to replace layout effect
• ci all green after fixes

11 eslint-disable reports for vibe-proof are introduced, mostly potential race conditions. These require future refactors.

re-ran codegen to update artifacts.

Checklist

• pnpm check:ci and pnpm test both pass (plus cargo fmt / clippy / test for Rust changes)
• I ran the affected surface and observed the change working
• If a wire message changed: WIRE_PROTOCOL_VERSION is bumped
• New code and assets are my own work, or their origin and license compatibility are noted above
• Docs and comments are updated where behavior changed

[greptile-apps]

Greptile Summary

This housekeeping PR resolves existing ESLint violations across the monorepo while preserving the application’s overall architecture.

• Reworks collection iteration, callback declarations, imports, and React effect/ref patterns to satisfy lint rules.
• Updates lint-related dependencies, shared utility packages, workspace manifests, and generated lockfile artifacts.
• Refreshes associated tests and generated assets.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains.

No blocking failure remains.

Important Files Changed

Filename	Overview
apps/desktop/src/renderer/src/shell/desktop-shell.tsx	Replaces layout-effect-driven panel host attachment with stable callback refs; no eligible follow-up defect was established.
packages/presentation/ui/src/keyboard/hooks.ts	Refactors keyboard listener and registration hooks for lint compliance; no publishable follow-up issue was established.
packages/foundation/schema/src/remote-config.ts	Applies utility and iteration refactors to remote configuration handling without an established contract regression.
packages/host/agent-adapter/src/native/claude-code.ts	Performs extensive lint-oriented control-flow and iteration cleanup in the Claude Code adapter.
pnpm-lock.yaml	Regenerates dependency resolution after workspace-wide utility and lint-tool upgrades.

_{Reviews (4): Last reviewed commit: "Update packages/host/agent-adapter/src/n..." | Re-trigger Greptile}

[pullfrog]

ℹ️ No correctness defects found. Two process concerns (undeclared dependency scope, one release-age pin) and a few nitpicks.

Reviewed changes

• A repo-wide ESLint codemod across 340 files: ~500 for...of → hoisted-len indexed loops, Object.fromEntries → reduce object builders, [...x].map() → Array.from(x, fn), localeCompare → module-scope Intl.Collator, setTimeout(() => f(a), ms) → setTimeout(f, ms, a), codePointAt → charCodeAt, and the sukka/* → vibe-proof/* rule-id namespace migration.
• eslint-suppressions.json (ESLint native bulk suppressions) is deleted outright.
• ~15 new split0th / split1st call sites from foxts/split-nth, replacing str.split(sep, n)[i], including two request-routing hot paths (transport/src/http-server.ts, daemon/src/sim/mcp-endpoint.ts).
• Genuine behavioral rewrites folded into the sweep: desktop-shell.tsx moves from useState + useLayoutEffect DOM capture to React 19 callback refs; transport/src/hub.ts converts an if/else if chain to switch (p.kind); keyboard/hooks.ts adopts foxact/use-abortable-effect with { signal } listener teardown; brand-assets.ts / verify-artifacts switch to lstatSync(…, { throwIfNoEntry: false }) with an explicit undefined guard.
• Dependency and toolchain bumps: foxts → ^5.9.1 across 8 packages, eslint ^10.8.0 → ^10.9.1, the sukka config/react/formatter trio ^8.14.2 → ^9.0.1 (major), packageManager pnpm@11.10.0 → pnpm@11.24.0, verkit@0.3.2 added, and ~2,874 lockfile lines.

✅ What I verified rather than assumed

Three properties of a codemod this size are only trustworthy if actually checked, so I checked them:

• Hoisted len never freezes a live worklist. Every one of the ~500 converted loops was classified by whether its body can grow the array it iterates. None can. The one genuine BFS worklist — packages/host/engine/src/workspace/file-suggest-service.ts — is correctly exempted with a prefer-indexed-array-loop disable and a comment explaining why both .length reads must stay live. Good catch by the author; that's the case that would have silently truncated file search.
• foxts/split-nth is an exact substitution. split0th(s, sep) ≡ s.split(sep)[0] and split1st(s, sep) ≡ s.split(sep)[1] — split1st returns the segment between the first and second separator, not the remainder, so split1st(name, '/') really does match name.split('/', 2)[1] for scoped package names in host/assets/src/closure.ts. (Worth stating explicitly: had it returned the remainder, the npm tarball URL builder and the MIME-subtype label would both have been wrong.)
• The desktop-shell.tsx callback refs are not a regression. The docked and maximized panel regions can never both render the ref'd <div>: contentHidden gates it, and the two instances are driven by the same expandedPanel value in layout/workspace.tsx:280-292. The new attachRightContentHost / attachBottomContentHost also ignore the null call entirely and just re-parent a useSingleton-owned host node, so there is no state to clobber. This is strictly simpler than the useState + useLayoutEffect version it replaces.

Also confirmed: OPENCODE_TOOL_NAME_SANITIZE_RE carries the g flag, so the replace → replaceAll change in native/opencode/history.ts cannot throw; and providers/routing.ts exports only a type, so export * → export type * in presentation/ui/src/shell/index.ts is safe.

⚠️ The PR body under-describes what's in the commit

The title and description read as a pure lint sweep, but the diff also carries a major-version bump of the ESLint config trio (eslint-config-sukka, @eslint-sukka/react, eslint-formatter-sukka ^8.14.2 → ^9.0.1), a packageManager change from pnpm@11.10.0 to pnpm@11.24.0, a new verkit@0.3.2 dependency, and hand-edited version ranges in packages/vendor/coss-ui/package.json (@base-ui/react ^1.6.0 → ^1.7.0, foxact ^0.3.8 → ^0.3.10, input-otp ^1.4.2 → ^1.5.0).

Two things follow from that:

1. The sukka major bump is almost certainly why the codemod exists — new rules landed. Saying so in the body turns 340 files of unexplained churn into a legible consequence of one line.
2. packages/vendor/coss-ui is the repo's designated vendored package: AGENTS.md says upstream formatting is "preserved as-synced — never hand-edit". Bumping its dependency ranges directly is a divergence from that sync contract. If these came from an upstream coss-ui resync, that's fine and worth stating; if they were bumped by hand to satisfy the lockfile, it will get clobbered on the next sync.

Neither blocks the PR — but a reviewer skimming "make eslint happy" will not expect to be reviewing a toolchain upgrade.

ℹ️ foxact@0.3.10 may trip pnpm's release-age gate

.github/dependabot.yml:6 documents that pnpm 11's default minimumReleaseAge (1 day) gates installs, and pnpm-workspace.yaml carries a 90-entry minimumReleaseAgeExclude allowlist for exactly this. Of the newly pinned versions:

• foxts — allow-listed (bare entry, line 141) ✅
• eslint-config-sukka / @eslint-sukka/react / eslint-formatter-sukka — covered by the eslint-*-sukka and @eslint-sukka/* globs ✅
• eslint@10.9.1 (2026-08-24), input-otp@1.5.0 (2026-08-18), verkit@0.3.2 (2026-08-15), @base-ui/react@1.7.0 (2026-08-04) — all comfortably past 1 day ✅
• foxact@0.3.10 — published 2026-08-27T16:19Z, and not in the allowlist.

That is right on the 24-hour boundary as of this PR. A CI runner doing a cold install before 2026-08-27T16:19Z + 24h would be refused the version the lockfile pins. It self-heals with the clock, so this is informational rather than blocking — but if CI has already gone red on install, this is the reason.

ℹ️ Nitpicks

• Object.defineProperty is applied unevenly. Eight of the new reduce-based object builders use Object.defineProperty with a comment explaining that acc.__proto__ = … would mutate the prototype instead of creating an own property — remote-config.ts even notes that a JSON __proto__ member "must survive into the canonical payload". Four sites use plain acc[key] = value instead. Three of those four take keys from in-repo string literals or a closed Zod enum and are unreachable; the fourth is flagged inline below. Not a vulnerability, but the rationale the author wrote eight times should either apply everywhere or the safe sites should say why they're special.
• The codemod emits a redundant alias in ~10 files: const index = i, item = items[i]; where i is already the loop variable (chat/activity-groups.ts, foundation/common/src/config/i-json.ts, node/release-artifact.ts, node/release-compliance.ts, schema/src/model/{linkcode-marketplace,permission,question}.ts, remote-config.ts, native/claude-code.ts, native/opencode/adapter.ts, engine/src/__tests__/engine-sessions.test.ts, workbench/src/history/use-provider-history.ts). Naming the loop variable index directly drops a line at each site.
• presentation/ui/src/lib/base-ui.ts drops import type * as React from 'react' while the file still writes React.SyntheticEvent, so the type now resolves through the @types/react UMD global namespace rather than an explicit import. It typechecks, but it's a silent dependency on ambient types in a package that otherwise imports what it uses.
• shell/git/diff-viewer.tsx hoists DiffViewerSkeleton from a component to a module-scope diffViewerSkeletonJsx element. Correct (React permits reusing element objects), and it does remove a component identity — just noting it's a rendering-semantics change, not a rename.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

[pullfrog]

ℹ️ All four of my earlier comments are genuinely addressed — I've replied to and resolved those threads. One informational note below, plus a correction to a call I got wrong last round.

Three new commits since my last review (96df5ef7..e52164a4, 16 files, +92/−89).

0290143b is a real bug fix, and a good catch. Resolving artifact paths against realpath(root) matters because a symlinked root — macOS /var → /private/var — makes every legitimate path read as an escape. Containment still holds for absolute and .. inputs, so the fix doesn't weaken the guard. CI never caught this because /tmp on Linux isn't a symlink and the tests use bare mkdtemp(join(tmpdir(), …)); codex-commands.test.ts:369 shows the repo already wraps with realpath elsewhere. I grepped every realpath( site — no sibling instance of the same bug.

b96a69d8 is faithful to the lockfile. I ran the repo's own drift guard (closure.test.ts → "the committed pi manifest matches the lockfile") and cross-read pnpm-lock.yaml: @google/genai@1.52.0 and @mistralai/mistralai@2.2.6 both now list ws: 8.21.1, matching the hoisted top-level, so the nested copies correctly disappear. See the one scope question below.

Tests run locally, all passing: foundation/{schema,common} + host/agent-adapter + presentation/ui (148 files / 1537 tests), closure.test.ts + common/src/node/__tests__ (7 / 59), engine-sessions.test.ts + client/workbench (43 / 260).

ℹ️ The pi closure revision change forces every existing user to re-download the closure

packages/host/assets/src/pi-closure.gen.ts moves revision: 'd6ff0228b84fa10c' → '6fdb7aeb2039774d' while pi itself stays at 0.80.6. Because installIdentity keys the store directory on <version>+<revision>:

// packages/host/assets/src/install.ts:50
return isClosureDescriptor(descriptor) ? `${version}+${descriptor.closure.revision}` : version;

…the new identity is 0.80.6+6fdb7aeb2039774d, a directory nobody has. Every existing pi user re-fetches the whole ~230-package closure.

The manifest is correct — it's the honest output of the generator against the current lockfile. But the delta is entirely incidental transitive churn from the pnpm 11.24 lockfile regen, not anything about pi: ws@8.21.0 nested copies deduped away, minimatch@10.2.5 newly nested under pi (top-level moved 10.2.5→10.2.6), @opentelemetry/semantic-conventions 1.41.1→1.43.0, brace-expansion 5.0.8→5.0.9, has-flag@4.0.0 and supports-color@8.1.1 added.

Not a defect, and nothing to change in the code — just flagging that a lint-sweep PR carries a full re-download for every pi user, which is a scope call only you can make. Committing the regen is clearly right if the lockfile bump is staying in this PR; the question is whether the lockfile bump belongs here at all (which is the same scope-creep thread from my last review, still open).

Not re-raising

The foxact@0.3.10 release-age gate from my last review is moot — 0.3.10 is now well past the 24h minimumReleaseAge boundary. The three items I verified in round 1 (hoisted len never freezing a live worklist, foxts/split-nth being an exact substitution, desktop-shell.tsx callback refs) are unaffected by these commits and stay settled.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}