Beautiful, AI-native markdown IDE and LLM wiki
TypeScript 2nd-brainagent-skillsclaudecodexcompany-braindocsknowledge-knowledge-baseknowledge-managementllmllm-wikillm-wiki-karpathymarkdownmarkdown-editormdmdxnotespkmsecond-brainskills
chore(open-knowledge): test-file naming convention + CI guards (#2323)
Nick Gomezde0c3240a35e
diff --git a/biome-plugins/README.md b/biome-plugins/README.mdindex f1dd2f58d..e9f7bc143 100644--- a/biome-plugins/README.md+++ b/biome-plugins/README.md@@ -1,6 +1,6 @@# Biome GritQL plugins-Custom lint rules for this workspace, registered in [`biome.jsonc`](../biome.jsonc) at the top-level `plugins` array (workspace-wide) OR a scoped `overrides[].plugins` entry (file-specific — used when the rule's invariant only applies to a known subset of files; see `playwright-topass-budget.grit` below). Each `.grit` file is a single GritQL pattern (or `or { ... }` of patterns) emitting diagnostics via `register_diagnostic()`.+Custom lint rules for this workspace, registered in [`biome.jsonc`](../biome.jsonc) at the top-level `plugins` array (workspace-wide) OR a scoped `overrides[].plugins` entry (file-specific — used when the rule's invariant only applies to a known subset of files). Each `.grit` file is a single GritQL pattern (or `or { ... }` of patterns) emitting diagnostics via `register_diagnostic()`.Plugins surface as lint errors during `biome check` (i.e. `bun run lint` and `bun run check`) and as inline editor squiggles via the Biome LSP.@@ -84,21 +84,6 @@ createPortal(Plugin: [`biome-plugins/no-unportaled-editor-content.grit`](no-unportaled-editor-content.grit). Fixture: [`biome-plugins/__fixtures__/no-unportaled-editor-content.fixture.tsx`](__fixtures__/no-unportaled-editor-content.fixture.tsx). Test: [`packages/app/tests/integration/no-unportaled-editor-content.test.ts`](../packages/app/tests/integration/no-unportaled-editor-content.test.ts). See [PRECEDENTS.md #44](../PRECEDENTS.md) for the H6 cross-doc DOM bleed contract and [PRECEDENTS.md #42](../PRECEDENTS.md#custom-lint-enforcement-precedent-42) for the GritQL-plugin convention.-### `playwright-topass-budget.grit`--Flags `toPass({ timeout: N })` calls where `N` is a thousand-range literal below `15_000` (the canonical range `1_000`–`14_999`, with or without the `_` digit separator). The rule pairs with `tests/smoke/calibration.test.ts` Invariant B: the calibration unit test enforces the boundary at `bun run check` time (PR-tier gate); this plugin fires at editor-save / lint-time so authors see the regression instantly.--**Why 15s.** macOS `open(1)` Apple-Event delivery + window creation + IPC roundtrip empirically takes 2-8s on a healthy CI runner and can hit 8.8s under sustained load. 15s gives 2-3x headroom over typical worst case. Precedent: `external-link.e2e.ts:83` `firstWindow({ timeout: 15_000 })`.--**Scoped via `overrides[].plugins`** to `packages/desktop/tests/smoke/{deep-link,external-link}.e2e.ts` + the fixture itself. `packages/app`'s `*.e2e.ts` files drive Chromium in-process (NOT macOS `open(1)` Apple-Event dispatch), so their sub-15s `toPass` budgets are legitimate and must not be flagged. The override scope mirrors `tests/smoke/calibration.test.ts`'s `TOPASS_BUDGET_FILES` whitelist — same boundary, two enforcement layers (lint-time + check-time).--The rule does NOT catch:-- Sub-1000ms literals (e.g., `timeout: 500`) — the canonical range starts at `1_000`. Sub-1s budgets aren't realistic for the macOS Apple-Event path. Calibration unit test backstops all values regardless of magnitude.-- Non-literal expressions (variables, computed values, member access) — pattern matches numeric literals only. Calibration unit test reads the same regex contract from source text and would still catch non-literal forms via the parser's literal extraction.-- Invariant A (cumulative inner timeouts ≤ outer per-test budget) — requires aggregation across same-file helper-call graphs, which GritQL pattern matching can't express. Enforced exclusively by the calibration unit test.--Plugin: [`biome-plugins/playwright-topass-budget.grit`](playwright-topass-budget.grit). Fixture: [`biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx`](__fixtures__/playwright-topass-budget.fixture.tsx). Test: [`packages/desktop/tests/integration/playwright-topass-budget.test.ts`](../packages/desktop/tests/integration/playwright-topass-budget.test.ts). See [PRECEDENTS.md #42](../PRECEDENTS.md#custom-lint-enforcement-precedent-42) for the GritQL-plugin convention.-### `path-conditional-map-driven-origin.grit`Observer A origin discipline. Inside `packages/server/src/server-observers.ts`, every `Y.Doc.transact()` call MUST pass the sanctioned origin `OBSERVER_SYNC_ORIGIN` as its second argument (`doc.transact(fn, OBSERVER_SYNC_ORIGIN)`). Bare `doc.transact(fn)` - or a wrong origin - routes the write to `openknowledge-service` and breaks per-session UndoManager attribution (the `trackedOrigins` Set-identity match skips the transaction).@@ -162,7 +147,7 @@ Plugin: [`biome-plugins/playwright-prefer-to-have-count.grit`](playwright-prefer### `no-roundtrip-identity-oracle.grit`-Forbids the byte-fidelity round-trip oracle in public-mirrored tests. Asserting that re-serializing a freshly-parsed document yields back the *same* input — `serialize(parse(x))` (or the MarkdownManager method form `m.serialize(m.parse(x))`) compared equal to that same `x` via `.toBe` / `.toEqual` / `.toStrictEqual` or `===` — is the engine's byte-identity correctness oracle. That oracle is private quality-bar IP that the engine fidelity suite owns; this rule keeps a new public test from reintroducing it on the public mirror.+Forbids the byte-fidelity round-trip oracle in public-mirrored tests. Asserting that re-serializing a freshly-parsed document yields back the *same* input — `serialize(parse(x))` (or the MarkdownManager method form `m.serialize(m.parse(x))`) compared equal to that same `x` via `.toBe` / `.toEqual` / `.toStrictEqual` or `===` — is the engine's byte-identity correctness oracle, exercised by its own fidelity suite. A public test should pin a specific expected output instead; this rule keeps a new public test from reintroducing the general oracle.**Identity, not contract.** The rule fires only when the parse input and the expected value are the *same expression* — GritQL metavariable reuse (`$x` … `$x`) enforces textual equality. That is what separates the oracle from the assertions that must stay public and green:@@ -170,7 +155,7 @@ Forbids the byte-fidelity round-trip oracle in public-mirrored tests. Asserting- The **Bridge-invariant comparator** `normalizeBridge(a) === normalizeBridge(b)` (precedent #38, the documented public contract) contains no `serialize(parse(...))` and is never flagged.- The **normalizing-construct detector** `serialize(parse(x)) !== x` uses `!==`, a different operator, and is never flagged.-**Scoped via `overrides[].plugins`** to the public-mirrored test surface (`packages/**/*.test.ts`, `*.test.tsx`, `*.e2e.ts`). The engine-test clusters that legitimately own the oracle are excluded as negative globs, mirroring the set excluded from the public mirror: `packages/app/tests/fidelity/**`, `packages/core/src/markdown/**/*.test.ts`, `packages/core/src/bridge/**/*.test.ts`, the enumerated byte-oracle tests, and the private `packages/md-conformance/**` estate. Inside those clusters the identity oracle is the whole point; outside them (the public surface) it is forbidden.+**Scoped via `overrides[].plugins`** to the public-mirrored test surface (`packages/**/*.test.ts`, `*.test.tsx`, `*.e2e.ts`). The engine-test clusters that legitimately own the oracle are excluded as negative globs: `packages/app/tests/fidelity/**`, `packages/core/src/markdown/**/*.test.ts`, `packages/core/src/bridge/**/*.test.ts`, the enumerated byte-oracle tests, and the `packages/md-conformance/**` estate. Inside those clusters the identity oracle is the whole point; outside them (the public surface) it is forbidden.The rule does NOT catch:- Round-trip identity through a helper (`mdRoundTrip(x)`, `normalize(...)`) or an intermediate variable (`const out = serialize(parse(x)); expect(out).toBe(x)`) — the pattern matches the inline call shape, not helper bodies or cross-statement data flow. Those forms live in the path-excluded fidelity suite, covered by exclusion.@@ -179,6 +164,26 @@ The rule does NOT catch:Plugin: [`biome-plugins/no-roundtrip-identity-oracle.grit`](no-roundtrip-identity-oracle.grit). Fixture: [`biome-plugins/__fixtures__/no-roundtrip-identity-oracle.fixture.tsx`](__fixtures__/no-roundtrip-identity-oracle.fixture.tsx). Test: [`packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.ts`](../packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.ts). See [PRECEDENTS.md #42](../PRECEDENTS.md#custom-lint-enforcement-precedent-42) for the GritQL-plugin convention and [PRECEDENTS.md #38](../PRECEDENTS.md) for the Bridge-invariant contract.+### `no-inline-tolerance-class.grit`++Forbids a public-mirrored test from writing a bridge normalization-class value inline as a string literal. `BRIDGE_TOLERANCE_CLASSES` (`packages/core/src/bridge/normalize.ts`) is the bridge normalizer's catalog of byte-difference equivalence classes it tolerates. A public test should assert observable `normalizeBridge` equivalence between inputs rather than hard-coding one of those class labels inline — the label is an internal classification detail, and pinning it inline both couples the test to that detail and re-declares the catalog outside the modules that own it. `check-mirror-test-policy` Check B already blocks a public test from *importing* the catalog symbol; this rule closes the complementary gap where a test re-encodes a class value inline (`expect(applied).toBe('jsx-container-boundary-blank')`, an array of class names), past the import check.++**Identity, not substring.** The `or {}` matches a string-literal node whose value *is* exactly a catalog member, so the names that appear legitimately on the public surface as prose are not flagged:++- A class name inside a longer **test-title sentence** (`test('… (block-separator-collapse class)', …)`) — a different node value, so it does not fire.+- A class name embedded in a **docName** with a prefix (`'fr34-doc-start-thematic'`) — likewise a substring, not the whole value.+- A class name in a **comment** — GritQL matches the string-literal node, not trivia.++The match is quote-style independent (a single-quoted pattern matches the double-quoted form Biome emits). The four **universal text-encoding** classes — `bom`, `crlf`, `trailing-whitespace`, `trailing-newline` — are deliberately NOT matched: they are normalizations every text tool performs, not distinctive classes, and the public floor telemetry runtime (`tolerance-telemetry.ts`) surfaces them, so public tests legitimately assert that runtime emits `class: 'crlf'` for a CRLF input. The 12 markdown-fidelity classes plus those 4 universal classes partition the catalog exactly, and the fixture test's drift canary pins that partition — a class added to `BRIDGE_TOLERANCE_CLASSES` reddens until it is classified into one bucket.++**Scoped via `overrides[].plugins`** to the public-mirrored test surface (`packages/**/*.test.ts`, `*.test.tsx`, `*.e2e.ts`), with the catalog-owning clusters excluded as negative globs (the bridge tests, the markdown precision suite, `packages/app/tests/fidelity/**`, the enumerated byte-oracle one-offs, and `packages/md-conformance/**`), so the catalog stays usable in those clusters but the inline form is forbidden on the public surface.++The rule does NOT catch:+- A class name built by concatenation or template interpolation (`'doc-start-' + 'thematic'`) — neither operand is the whole value.+- A class name in a template literal — the pattern matches string-literal nodes, not template content.++Plugin: [`biome-plugins/no-inline-tolerance-class.grit`](no-inline-tolerance-class.grit). Fixture: [`biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsx`](__fixtures__/no-inline-tolerance-class.fixture.tsx). Test: [`packages/app/tests/lint-plugins/no-inline-tolerance-class.test.ts`](../packages/app/tests/lint-plugins/no-inline-tolerance-class.test.ts). See [PRECEDENTS.md #42](../PRECEDENTS.md#custom-lint-enforcement-precedent-42) for the GritQL-plugin convention.+## SuppressionInline `// biome-ignore` comments silence individual diagnostics. The most specific form names the rule and the reason:@@ -199,8 +204,8 @@ Current production suppressions:- `no-loosely-typed-webcontents-ipc`: 15 sites (`preload/index.ts` ×12, `shared/ipc-send.ts` ×1, `tests/smoke/theme-sync.e2e.ts` ×2)- `no-raw-html-interactive-element`: 20 file-level `biome-ignore-all` headers in `packages/app/src/{components,presence}/**` (pre-rule backlog awaiting shadcn migration; see the rule's section above for the ratchet contract)- `no-resolved-value-theme-source`: 0 sites-- `playwright-topass-budget`: 0 sites- `no-roundtrip-identity-oracle`: 0 sites+- `no-inline-tolerance-class`: 0 sites## Adding a new plugin@@ -235,7 +240,7 @@ language jsPick the scope:- **Workspace-wide** (default — used by `microcopy-ellipsis`, `no-loosely-typed-webcontents-ipc`, `no-resolved-value-theme-source`): add the path to the top-level `plugins` array. The rule fires on every linted file.-- **Scoped to specific files** (used by `playwright-topass-budget`): add an entry to the `overrides` array with `includes: [...]` listing the in-scope files (and the fixture path so the fixture test still triggers the rule) and `plugins: ['./biome-plugins/<rule-name>.grit']`. Use this when the rule's invariant only holds for a known subset of files — e.g., a budget that depends on a platform-specific code path. Document the scope-discipline rationale in the rule's docstring and assert it in the fixture-file test.+- **Scoped to specific files** (used by `no-raw-html-interactive-element`): add an entry to the `overrides` array with `includes: [...]` listing the in-scope files (and the fixture path so the fixture test still triggers the rule) and `plugins: ['./biome-plugins/<rule-name>.grit']`. Use this when the rule's invariant only holds for a known subset of files — e.g., a rule that should only fire under a specific source subtree. Document the scope-discipline rationale in the rule's docstring and assert it in the fixture-file test.Either shape participates in the same `biome check` pass; the override form just adds Biome's path matcher in front of the GritQL pattern.@@ -291,7 +296,7 @@ describe('<rule-name> GritQL plugin', () => {The "plugin is registered" test catches the failure mode where a `.grit` file is added but the `biome.jsonc#plugins` entry is missing.-**For override-scoped plugins** (step 2 second variant): swap the registration assertion for one that asserts the plugin is in `config.overrides[].plugins`, the matching override's `includes` covers every in-scope file (including the fixture), and the plugin is NOT at root `plugins[]` (so an accidental move from override to root, which would over-fire, fails). `playwright-topass-budget.test.ts` is the reference shape.+**For override-scoped plugins** (step 2 second variant): swap the registration assertion for one that asserts the plugin is in `config.overrides[].plugins`, the matching override's `includes` covers every in-scope file (including the fixture), and the plugin is NOT at root `plugins[]` (so an accidental move from override to root, which would over-fire, fails). `no-roundtrip-identity-oracle.test.ts` is the reference shape.### 5. Verify@@ -320,7 +325,7 @@ Add a section under `## Rules` with: what it flags, what it doesn't catch, links## Out of scope- **Autofix.** Biome 2.4's GritQL plugins are diagnostic-only. Plugin diagnostics cannot apply code fixes. If autofix is required, a different enforcement mechanism is needed (build-time codemod, separate `--fix` script).-- **GritQL-internal path filters.** GritQL itself doesn't support file-path allowlists. The natural scope of the GritQL pattern (e.g., JSX-only) is the primary in-pattern mechanism for excluding files; inline `// biome-ignore` comments handle the residual. When a rule needs explicit per-file scoping, register the plugin under Biome's `overrides[].plugins` instead (see `playwright-topass-budget` and step 2 of "Adding a new plugin") — that runs the path matcher at the Biome layer before invoking the GritQL pattern.+- **GritQL-internal path filters.** GritQL itself doesn't support file-path allowlists. The natural scope of the GritQL pattern (e.g., JSX-only) is the primary in-pattern mechanism for excluding files; inline `// biome-ignore` comments handle the residual. When a rule needs explicit per-file scoping, register the plugin under Biome's `overrides[].plugins` instead (see `no-raw-html-interactive-element` and step 2 of "Adding a new plugin") — that runs the path matcher at the Biome layer before invoking the GritQL pattern.- **CLI string content.** `process.stderr.write('...')` / `console.log` template-literal content is not reliably matchable via GritQL call-expression patterns (false-positive rate too high). Review discipline covers these surfaces.## Referencesdiff --git a/biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsx b/biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsxnew file mode 100644index 000000000..05dd74f96--- /dev/null+++ b/biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsx@@ -0,0 +1,68 @@+/**+ * Fixture for `no-inline-tolerance-class.grit`.+ *+ * Pairs 8 positive cases (a bridge normalization-class value+ * written inline as a string literal — plugin MUST fire) with negative cases+ * (the four universal text-encoding classes used as legitimate inline literals,+ * a class name appearing only as a SUBSTRING of test-title / docName prose, a+ * comment mention, a concatenation, and a non-catalog literal — plugin must NOT+ * fire). The fixture-file test asserts the diagnostic count with exact equality+ * (`toBe(8)`) so both a weakened pattern (drops below 8) and a widened pattern+ * that catches a negative (rises above 8) fail the gate.+ *+ * The universal-encoding negatives (`crlf`, `bom`, `trailing-whitespace`,+ * `trailing-newline`) are the load-bearing precision boundary: they are real+ * `BRIDGE_TOLERANCE_CLASSES` members, yet they are deliberately NOT flagged+ * because the public floor telemetry runtime surfaces them and public tests+ * legitimately assert that behavior. Flagging them would redden those floor+ * tests; the fixture pins that they stay silent.+ *+ * Deliberately NOT linted by the main `bun run lint` pass (biome-plugins/ is+ * outside the lint paths); only the scoped override in biome.jsonc reaches it,+ * via the fixture-file test.+ */++declare const expect: (v: unknown) => { toBe: (v: unknown) => void };+declare const applied: string;+declare const cls: string;++function positives() {+ // P1: a single distinctive fidelity class written inline.+ const single = 'emphasis-around-code';+ // P2-P5: four class labels written inline as an array.+ const reencoded = [+ 'doc-start-thematic',+ 'block-separator-collapse',+ 'table-align-row-spacing',+ 'row-no-trailing-pipe',+ ];+ // P6: asserting an observed class against an exact fidelity literal.+ expect(applied).toBe('jsx-container-boundary-blank');+ // P7: an equality comparison against an exact fidelity literal.+ const isCanonical = cls === 'list-indent-canonical';+ // P8: a fidelity class as an object property value.+ const record = { className: 'ordered-list-marker-number' };+ return { single, reencoded, isCanonical, record };+}++function negatives() {+ // N1-N4: the four universal text-encoding classes are real catalog members but+ // are deliberately not flagged — the public floor telemetry runtime emits+ // them and public tests assert that behavior. They must stay silent.+ const isCrlf = cls === 'crlf';+ const isBom = cls === 'bom';+ const fired = ['trailing-whitespace', 'trailing-newline'];+ // N5: a class name appearing only as a SUBSTRING of a test-title sentence —+ // the rule keys on the literal's whole value, not a substring.+ const title = 'watchdog tolerates a doc-start-thematic divergence at column 1';+ // N6: a class name embedded as a substring of a docName carrying a prefix.+ const docName = 'fr34-doc-start-thematic';+ // N7: a class name mentioned in a comment (emphasis-around-code) is trivia,+ // not a string-literal node; this assignment is an ordinary literal.+ const mode = 'replace';+ // N8: split across a concatenation — neither operand is the whole value.+ const split = `doc-start-${'thematic'}`;+ return { isCrlf, isBom, fired, title, docName, mode, split };+}++export { negatives, positives };diff --git a/biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx b/biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsxdeleted file mode 100644index c9740d6c3..000000000--- a/biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx+++ /dev/null@@ -1,50 +0,0 @@-// FIXTURE — drives `playwright-topass-budget.test.ts` via shell-out to-// `biome check`. Not part of the main lint (lives outside the lint-// command's path list).-//-// Five positive cases (deliberate violations — plugin must fire) + four-// negative cases (well-calibrated `toPass` budgets — plugin must NOT-// fire). Exact-equality (`toBe(5)`) in the test catches both-// false-negative regressions (drop below 5) and false-positive widenings-// (above 5).--// We're not actually executing this fixture — it just needs to parse-// with realistic call shapes. `expect.toPass` is the real shape.-declare const expect: { (fn: () => unknown): { toPass: (opts: { timeout: number }) => unknown } };--// === Positive cases — toPass budgets BELOW 15_000 ===--// (1) Canonical 5s budget — the exact shape PR #533 originally shipped-// for deep-link/external-link tests.-void expect(() => 'x').toPass({ timeout: 5_000 });--// (2) Same value without the underscore separator — both forms are-// valid JS literals; the plugin must catch both.-void expect(() => 'x').toPass({ timeout: 5000 });--// (3) 10s budget — also below 15s threshold. Tests the 10000-14999 band.-void expect(() => 'x').toPass({ timeout: 10_000 });--// (4) Edge case just below threshold — 14_999 is the largest still-flagged value.-void expect(() => 'x').toPass({ timeout: 14_999 });--// (5) Sub-second budget — way too short for Apple-Event roundtrip,-// plugin must flag.-void expect(() => 'x').toPass({ timeout: 1_000 });--// === Negative cases — toPass budgets AT OR ABOVE 15_000 ===--// (6) Exact threshold — must NOT fire. The calibration unit test's-// invariant is `>= 15_000`; the plugin mirrors that boundary.-void expect(() => 'x').toPass({ timeout: 15_000 });--// (7) Comfortable headroom — the value the fix in this PR adopts.-void expect(() => 'x').toPass({ timeout: 30_000 });--// (8) `15000` without underscore — the literal form variant must NOT fire.-void expect(() => 'x').toPass({ timeout: 15000 });--// (9) `toPass()` without options — no `timeout` to flag, must NOT fire.-// Playwright's default ~5s applies internally, but that's a separate-// concern; the plugin only checks declared `timeout:` literals.-void expect(() => 'x').toPass();diff --git a/biome-plugins/no-inline-tolerance-class.grit b/biome-plugins/no-inline-tolerance-class.gritnew file mode 100644index 000000000..97d35fe5a--- /dev/null+++ b/biome-plugins/no-inline-tolerance-class.grit@@ -0,0 +1,60 @@+// no-inline-tolerance-class — forbid an inline bridge normalization-class value in a public test.+//+// `BRIDGE_TOLERANCE_CLASSES` (packages/core/src/bridge/normalize.ts) is the+// bridge normalizer's catalog of byte-difference equivalence classes it tolerates.+// A public test should assert OBSERVABLE behavior — that two inputs reduce to the+// same `normalizeBridge` output — rather than hard-coding one of those class+// labels as an inline string literal: the label is an internal classification+// detail, and pinning it inline both couples the test to that detail and+// re-declares the catalog outside the modules that own it. `check-mirror-test-policy`+// Check B blocks a public test from IMPORTING the catalog symbol; this rule closes+// the complementary gap where a test re-encodes a class value as an inline string+// literal, past the import check.+//+// NOT matched, by design: the four universal text-encoding classes — `bom`,+// `crlf`, `trailing-whitespace`, `trailing-newline`. Those are normalizations+// every text tool performs, and the public floor telemetry runtime+// (packages/core/src/bridge/tolerance-telemetry.ts) surfaces them; public tests+// legitimately assert that runtime emits `class: 'crlf'` for a CRLF input.+// Matching them would redden those floor tests. The fixture test's drift canary+// pins this partition: every catalog member is either a matched arm here or one+// of those four universal classes, so a class added to the catalog reddens until+// it is classified.+//+// IDENTITY, not substring: the `or {}` matches a string-literal node whose value+// IS exactly a catalog member. A class name that appears only as a substring of a+// larger string — a test-title sentence (`'... (doc-start-thematic class)'`), a+// docName (`'fr34-doc-start-thematic'`), or a label — has a different node value+// and does not fire. The match is quote-style independent (a single-quoted+// pattern matches the double-quoted form Biome emits).+//+// Does NOT catch (documented honestly):+// - A class name built by concatenation (`'doc-start-' + 'thematic'`) or+// interpolation — neither operand is the whole value.+// - A class name in a template literal or a comment — GritQL matches the+// string-literal node, not template content or trivia.+//+// Suppress a legitimate inline use (none expected on the public surface) with:+// // biome-ignore lint/plugin/no-inline-tolerance-class: <reason>++language js++or {+ `'commonmark-escape'` as $lit,+ `'emphasis-around-code'` as $lit,+ `'leading-newline'` as $lit,+ `'doc-start-thematic'` as $lit,+ `'block-separator-collapse'` as $lit,+ `'table-align-row-spacing'` as $lit,+ `'row-no-trailing-pipe'` as $lit,+ `'list-indent-canonical'` as $lit,+ `'ordered-list-marker-number'` as $lit,+ `'paragraph-continuation-indent'` as $lit,+ `'jsx-container-boundary-blank'` as $lit,+ `'blank-line-collapse'` as $lit+} where {+ register_diagnostic(+ span = $lit,+ message = "Inline bridge normalization-class value in a public test. Assert observable `normalizeBridge` equivalence between inputs instead of hard-coding a BRIDGE_TOLERANCE_CLASSES label as a string literal. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#no-inline-tolerance-classgrit"+ )+}diff --git a/biome-plugins/no-roundtrip-identity-oracle.grit b/biome-plugins/no-roundtrip-identity-oracle.gritindex 8a2093bb7..11a75d2b0 100644--- a/biome-plugins/no-roundtrip-identity-oracle.grit+++ b/biome-plugins/no-roundtrip-identity-oracle.grit@@ -3,10 +3,10 @@// Asserting that re-serializing a freshly-parsed document yields back the// SAME input bytes — `serialize(parse(x))` (or the MarkdownManager method// equivalent `m.serialize(m.parse(x))`) compared equal to that same `x` — is-// the engine's byte-identity correctness oracle. That oracle is the private-// quality-bar IP and lives in the private engine fidelity suite; it must not-// ship on the public mirror. This rule flags the inline assertion shape so a-// new public test can't reintroduce it.+// the engine's byte-identity correctness oracle. This general round-trip-identity+// check is exercised by the engine's own fidelity suite; a public test should pin+// a specific expected output instead. This rule flags the inline assertion shape+// so a new public test can't reintroduce the general oracle.//// IDENTITY, not contract: the rule only fires when the parse input and the// expected value are the SAME expression (metavariable reuse `$x` ... `$x`@@ -22,8 +22,8 @@// - Round-trip identity through a helper (`mdRoundTrip(x)`, `normalize(...)`)// or via an intermediate variable (`const out = serialize(parse(x)); ...`)// — GritQL matches the inline call shape, not helper bodies or cross--// statement data flow. Those forms live in the private fidelity suite,-// which is path-excluded from the mirror, so they are covered there.+// statement data flow. Those forms live in the engine's fidelity suite and+// are covered there.// - Matchers other than `toBe` / `toEqual` / `toStrictEqual` and operators// other than `===` (e.g. a custom `assertByteIdentical` helper).//@@ -46,7 +46,7 @@ or {},register_diagnostic(span = $assertion,- message = "Byte-fidelity round-trip oracle in a public test: `serialize(parse(x))` asserted equal to the same input `x` exposes the engine's byte-identity correctness oracle on the public mirror. Keep round-trip-identity assertions in the private engine fidelity suite (packages/app/tests/fidelity or packages/core/src/markdown), or assert against a fixed expected literal for a public contract test. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#no-roundtrip-identity-oraclegrit"+ message = "Byte-fidelity round-trip oracle in a public test: `serialize(parse(x))` asserted equal to the same input `x`. This general round-trip-identity check belongs to the engine's fidelity suite (packages/app/tests/fidelity or packages/core/src/markdown); a public test should assert a fixed expected literal for a specific contract instead. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#no-roundtrip-identity-oraclegrit")},// serialize(parse($x)) === $x in either operand order (covers `assert(...)`,@@ -71,7 +71,7 @@ or {},register_diagnostic(span = $cmp,- message = "Byte-fidelity round-trip oracle in a public test: `serialize(parse(x))` asserted equal to the same input `x` exposes the engine's byte-identity correctness oracle on the public mirror. Keep round-trip-identity assertions in the private engine fidelity suite (packages/app/tests/fidelity or packages/core/src/markdown), or assert against a fixed expected literal for a public contract test. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#no-roundtrip-identity-oraclegrit"+ message = "Byte-fidelity round-trip oracle in a public test: `serialize(parse(x))` asserted equal to the same input `x`. This general round-trip-identity check belongs to the engine's fidelity suite (packages/app/tests/fidelity or packages/core/src/markdown); a public test should assert a fixed expected literal for a specific contract instead. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#no-roundtrip-identity-oraclegrit")}}diff --git a/biome-plugins/playwright-topass-budget.grit b/biome-plugins/playwright-topass-budget.gritdeleted file mode 100644index 1eb4fc1b7..000000000--- a/biome-plugins/playwright-topass-budget.grit+++ /dev/null@@ -1,48 +0,0 @@-// playwright-topass-budget — flag `toPass({ timeout: N })` with N < 15_000.-//-// Pairs with `tests/smoke/calibration.test.ts` Invariant B (per-assertion-// timeout calibration). The calibration unit test enforces the invariant-// at `bun run check` time (PR-tier gate); this rule fires at editor-save-// / lint-time so authors see the regression instantly instead of waiting-// for the gate to run.-//-// Why 15_000: macOS `open(1)` Apple-Event delivery + window creation +-// IPC roundtrip empirically takes 2-8s on a healthy CI runner and can hit-// 8.8s under sustained load. 15s gives 2-3x headroom over typical worst-// case. Precedent: `external-link.e2e.ts:83` `firstWindow({ timeout:-// 15_000 })`.-//-// Scope: the cumulative-budget invariant (Invariant A — cumulative inner-// timeouts ≤ outer per-test budget) is enforced by the calibration unit-// test, NOT here. Invariant A requires aggregation across same-file-// helper-call graphs, which GritQL pattern matching can't express. Only-// Invariant B's single-pattern check fits cleanly.-//-// Regex captures the canonical `N_000` / `N000` thousand-range literal-// forms: 1000-9999 (single-digit prefix + 3 digits) and 10000-14999-// (1[0-4] prefix + 3 digits), with optional `_` digit separator. Above-// 15_000 doesn't match — those are well-calibrated.-//-// Suppress legitimate uses (e.g., a fast non-Electron path that doesn't-// touch Apple-Event delivery) with:-// // biome-ignore lint/plugin/playwright-topass-budget: <reason>--language js--// `toPass` is always invoked as a member call on the result of `expect(...)`-// (per Playwright's API surface). The dominant shape is-// `expect(probeFn).toPass({ timeout: N })`. `$_.toPass($opts)` matches the-// member-call form and captures the options bag as `$opts`.-//-// `contains \`timeout: $n\`` traverses into the options-object literal to-// find the `timeout: N` property; `$n` captures the numeric-literal AST-// node, whose text the regex below matches against.--`$_.toPass($opts)` as $call where {- $opts <: contains `timeout: $n`,- $n <: r"^(?:[0-9]|1[0-4])_?[0-9]{3}$",- register_diagnostic(- span = $call,- message = "Invariant B: raise the `toPass({ timeout })` budget to at least 15_000ms — macOS Apple-Event / IPC roundtrip needs ≥15s headroom under CI runner-class load. See https://github.com/inkeep/open-knowledge/blob/main/biome-plugins/README.md#playwright-topass-budgetgrit"- )-}diff --git a/biome.jsonc b/biome.jsoncindex f6eb1713f..c597b6c3a 100644--- a/biome.jsonc+++ b/biome.jsonc@@ -111,21 +111,6 @@],"plugins": ["./biome-plugins/class-proof-registration-discipline.grit"]},- {- // Apple-Event / IPC roundtrip toPass budget enforcement is scoped to- // the desktop-smoke files (and the plugin's own fixture for testing).- // packages/app's *.e2e.ts files drive Chromium in-process, NOT macOS- // `open(1)` Apple-Event dispatch, so their sub-15s toPass budgets are- // legitimate (fast in-process roundtrip, no Launch Services latency).- // Pairs with `tests/smoke/calibration.test.ts` Invariant B, which is- // scoped to the same surface via `TOPASS_BUDGET_FILES`.- "includes": [- "packages/desktop/tests/smoke/deep-link.e2e.ts",- "packages/desktop/tests/smoke/external-link.e2e.ts",- "biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx"- ],- "plugins": ["./biome-plugins/playwright-topass-budget.grit"]- },{// One-shot `expect(await locator.count())` reads never retry and are// the canonical hidden-flake assertion shape (2026-06 e2e audit).@@ -219,11 +204,11 @@// Byte-fidelity round-trip oracle ban, scoped to the public-mirrored test// surface. `serialize(parse(x))` (or the MarkdownManager method form)// asserted equal to the same `x` is the engine's byte-identity correctness- // oracle — private quality-bar IP. It legitimately lives in the engine- // fidelity suite (packages/app/tests/fidelity), the markdown + bridge- // precision suites, and the enumerated byte-oracle tests; those are- // excluded from the public mirror, and are excluded here too so the oracle- // stays allowed there but forbidden on the public surface.+ // oracle. It legitimately lives in the engine fidelity suite+ // (packages/app/tests/fidelity), the markdown + bridge precision suites, and+ // the enumerated byte-oracle tests; those are excluded from this override's+ // scope so the oracle stays allowed there but forbidden on the public+ // surface.// packages/md-conformance is the private conformance estate// (also mirror-excluded). The fixture is in scope so its planted oracle// drives the fixture-file test; the Bridge-invariant comparator@@ -237,15 +222,35 @@"!packages/app/tests/fidelity/**","!packages/core/src/markdown/**/*.test.ts","!packages/core/src/bridge/**/*.test.ts",- "!packages/server/src/prd-6654-multi-client-repro.test.ts",- "!packages/app/tests/integration/source-mode-byte-preservation.test.ts",- "!packages/app/tests/integration/init-load-byte-stable.test.ts",- "!packages/app/tests/integration/init-load-byte-stable-corpus-coverage.test.ts",- "!packages/app/tests/stress/init-load-byte-stable.e2e.ts",- "!packages/app/tests/stress/single-file-ephemeral.e2e.ts",+ "!**/*.private.*","biome-plugins/__fixtures__/no-roundtrip-identity-oracle.fixture.tsx"],"plugins": ["./biome-plugins/no-roundtrip-identity-oracle.grit"]+ },+ {+ // Inline bridge normalization-class value ban, scoped to the+ // public-mirrored test surface. BRIDGE_TOLERANCE_CLASSES+ // (packages/core/src/bridge/normalize.ts) is the bridge normalizer's class+ // catalog; importing it into a public test is blocked by+ // check-mirror-test-policy Check B, and this rule blocks the complementary+ // inline re-encoding. The catalog legitimately lives in the bridge tests,+ // the markdown precision suite, the fidelity suite, and the byte-oracle+ // one-offs — excluded from this override's scope (as they are from the+ // plugin pass) so the catalog stays allowed there but the inline form is+ // forbidden on the public surface. The fixture is in scope so its planted+ // literals drive the fixture-file test.+ "includes": [+ "packages/**/*.test.ts",+ "packages/**/*.test.tsx",+ "packages/**/*.e2e.ts",+ "!packages/md-conformance/**",+ "!packages/app/tests/fidelity/**",+ "!packages/core/src/markdown/**/*.test.ts",+ "!packages/core/src/bridge/**/*.test.ts",+ "!**/*.private.*",+ "biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsx"+ ],+ "plugins": ["./biome-plugins/no-inline-tolerance-class.grit"]}],"files": {diff --git a/knip.config.ts b/knip.config.tsindex 020d6d598..662ca81cf 100644--- a/knip.config.ts+++ b/knip.config.ts@@ -52,6 +52,8 @@ export default {'scripts/assert-smoke-not-vacuous.mjs': ['files'],'scripts/assert-smoke-not-vacuous.test.mjs': ['files'],'scripts/check-no-ci-skip-in-e2e.test.mjs': ['files'],+ 'scripts/assert-app-built.mjs': ['files'],+ 'scripts/assert-app-built.test.mjs': ['files'],'.github/scripts/cla-gate.test.mjs': ['files'],'.github/scripts/cla-gate.mjs': ['exports'],'.github/scripts/bridge-public-pr-to-monorepo.test.mjs': ['files'],diff --git a/packages/app/tests/lint-plugins/no-inline-tolerance-class.test.ts b/packages/app/tests/lint-plugins/no-inline-tolerance-class.test.tsnew file mode 100644index 000000000..2a41625bc--- /dev/null+++ b/packages/app/tests/lint-plugins/no-inline-tolerance-class.test.ts@@ -0,0 +1,69 @@+import { describe, expect, test } from 'bun:test';+import { spawnSync } from 'node:child_process';+import { readFileSync } from 'node:fs';+import { join } from 'node:path';++const REPO_ROOT = join(__dirname, '..', '..', '..', '..');+const FIXTURE_REL = 'biome-plugins/__fixtures__/no-inline-tolerance-class.fixture.tsx';+const PLUGIN_REL = './biome-plugins/no-inline-tolerance-class.grit';+const GRIT_ABS = join(REPO_ROOT, 'biome-plugins/no-inline-tolerance-class.grit');+const CATALOG_SOURCE_ABS = join(REPO_ROOT, 'packages/core/src/bridge/normalize.ts');++describe('no-inline-tolerance-class GritQL plugin', () => {+ test('fires on exactly 8 inline fidelity-class literals (and on no negative case)', () => {+ const result = spawnSync('bunx', ['biome', 'check', FIXTURE_REL], {+ cwd: REPO_ROOT,+ encoding: 'utf-8',+ });+ expect(result.error).toBeUndefined();+ expect(result.status).not.toBe(0);+ const output = `${result.stdout}\n${result.stderr}`;+ const fires = (output.match(/Inline bridge normalization-class value in a public test/g) ?? [])+ .length;+ expect(fires).toBe(8);+ expect(output).toContain('hard-coding a BRIDGE_TOLERANCE_CLASSES label');+ expect(output).toMatch(/https?:\/\/[^\s]+/);+ expect(output).toContain('biome-plugins/README.md#no-inline-tolerance-classgrit');+ });++ test('plugin is registered as an override scoped to the public test surface (not workspace-wide)', () => {+ const config = require(join(REPO_ROOT, 'biome.jsonc'));+ const rootPlugins: string[] = config.plugins ?? [];+ expect(rootPlugins).not.toContain(PLUGIN_REL);++ const overrides: Array<{ includes?: string[]; plugins?: string[] }> = config.overrides ?? [];+ const entry = overrides.find((o) => (o.plugins ?? []).includes(PLUGIN_REL));+ expect(entry).toBeDefined();+ const includes = entry?.includes ?? [];+ expect(includes).toContain(FIXTURE_REL);+ for (const excluded of [+ '!packages/md-conformance/**',+ '!packages/app/tests/fidelity/**',+ '!packages/core/src/markdown/**/*.test.ts',+ '!packages/core/src/bridge/**/*.test.ts',+ '!**/*.private.*',+ ]) {+ expect(includes).toContain(excluded);+ }+ });++ test('matched fidelity set + universal-encoding set partition BRIDGE_TOLERANCE_CLASSES', () => {+ const UNIVERSAL_ENCODING = ['bom', 'crlf', 'trailing-whitespace', 'trailing-newline'];++ const catalogSrc = readFileSync(CATALOG_SOURCE_ABS, 'utf-8');+ const arrayBody = catalogSrc.match(/BRIDGE_TOLERANCE_CLASSES\s*=\s*\[([\s\S]*?)\]/)?.[1];+ expect(arrayBody).toBeDefined();+ const catalog = [...(arrayBody ?? '').matchAll(/'([^']+)'/g)].map((m) => m[1]).sort();+ expect(catalog.length).toBeGreaterThan(0);++ const gritArms = readFileSync(GRIT_ABS, 'utf-8')+ .split('\n')+ .filter((line) => !line.trimStart().startsWith('//'))+ .join('\n');+ const matched = [...gritArms.matchAll(/`'([^']+)'`/g)].map((m) => m[1]).sort();++ expect(matched.filter((c) => UNIVERSAL_ENCODING.includes(c))).toEqual([]);+ const union = [...new Set([...matched, ...UNIVERSAL_ENCODING])].sort();+ expect(union).toEqual(catalog);+ });+});diff --git a/packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.ts b/packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.tsindex 6246f2e30..0142e57b0 100644--- a/packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.ts+++ b/packages/app/tests/lint-plugins/no-roundtrip-identity-oracle.test.ts@@ -17,9 +17,7 @@ describe('no-roundtrip-identity-oracle GritQL plugin', () => {const output = `${result.stdout}\n${result.stderr}`;const fires = (output.match(/Byte-fidelity round-trip oracle in a public test/g) ?? []).length;expect(fires).toBe(10);- expect(output).toContain(- 'Keep round-trip-identity assertions in the private engine fidelity suite',- );+ expect(output).toContain('assert a fixed expected literal for a specific contract');expect(output).toMatch(/https?:\/\/[^\s]+/);expect(output).toContain('biome-plugins/README.md#no-roundtrip-identity-oraclegrit');});@@ -39,12 +37,7 @@ describe('no-roundtrip-identity-oracle GritQL plugin', () => {'!packages/app/tests/fidelity/**','!packages/core/src/markdown/**/*.test.ts','!packages/core/src/bridge/**/*.test.ts',- '!packages/server/src/prd-6654-multi-client-repro.test.ts',- '!packages/app/tests/integration/source-mode-byte-preservation.test.ts',- '!packages/app/tests/integration/init-load-byte-stable.test.ts',- '!packages/app/tests/integration/init-load-byte-stable-corpus-coverage.test.ts',- '!packages/app/tests/stress/init-load-byte-stable.e2e.ts',- '!packages/app/tests/stress/single-file-ephemeral.e2e.ts',+ '!**/*.private.*',]) {expect(includes).toContain(excluded);}diff --git a/packages/cli/src/commands/init.test.ts b/packages/cli/src/commands/init.test.tsindex 5671edbed..3cd7d96df 100644--- a/packages/cli/src/commands/init.test.ts+++ b/packages/cli/src/commands/init.test.ts@@ -52,6 +52,8 @@ import {} from './init.ts';import { LAUNCH_JSON_PORT } from './ui.ts';+const NATIVE_TOML_AVAILABLE = createTomlConfigEngine().backend === 'native';+describe('LAUNCH_UI_CHAIN_V1 (published launch.json recipe shell chain)', () => {it('is syntactically valid POSIX sh (sh -n)', () => {const result = spawnSync('sh', ['-n', '-c', LAUNCH_UI_CHAIN_V1], { encoding: 'utf-8' });@@ -447,7 +449,7 @@ describe('runInit', () => {expect(config.mcp_servers[result.editors[0].serverName]).toEqual(expectedDevMcpEntry());});- it('preserves existing Codex MCP entries', async () => {+ it.skipIf(!NATIVE_TOML_AVAILABLE)('preserves existing Codex MCP entries', async () => {mkdirSync(dirname(codexConfigPath()), { recursive: true });writeFileSync(codexConfigPath(),@@ -2131,18 +2133,21 @@ describe('classifyExistingMcpEntry', () => {});});- it('no-entry (not decline) on a valid Codex config with a 2^53+ integer', () => {- const path = resolveCodexConfigPath({ home: fakeHome, env: {} });- mkdirSync(dirname(path), { recursive: true });- writeFileSync(- path,- '# keep my comments\nmodel = "gpt-5"\n[mcp_servers.other]\ncommand = "node"\nstartup_timeout_ms = 9223372036854775807\n',- 'utf-8',- );- expect(classifyExistingMcpEntry(EDITOR_TARGETS.codex, '', fakeHome)).toEqual({- kind: 'no-entry',- });- });+ it.skipIf(!NATIVE_TOML_AVAILABLE)(+ 'no-entry (not decline) on a valid Codex config with a 2^53+ integer',+ () => {+ const path = resolveCodexConfigPath({ home: fakeHome, env: {} });+ mkdirSync(dirname(path), { recursive: true });+ writeFileSync(+ path,+ '# keep my comments\nmodel = "gpt-5"\n[mcp_servers.other]\ncommand = "node"\nstartup_timeout_ms = 9223372036854775807\n',+ 'utf-8',+ );+ expect(classifyExistingMcpEntry(EDITOR_TARGETS.codex, '', fakeHome)).toEqual({+ kind: 'no-entry',+ });+ },+ );it('present on a valid Codex config with a microsecond datetime and OK entry', () => {const path = resolveCodexConfigPath({ home: fakeHome, env: {} });diff --git a/packages/desktop/src/main/mcp-wiring.test.ts b/packages/desktop/src/main/mcp-wiring.test.tsindex 74fdfcfd8..cd7c1c437 100644--- a/packages/desktop/src/main/mcp-wiring.test.ts+++ b/packages/desktop/src/main/mcp-wiring.test.ts@@ -35,6 +35,8 @@ import {writeMcpStatusMarker,} from './mcp-wiring.ts';+const NATIVE_TOML_AVAILABLE = createTomlConfigEngine().backend === 'native';+function memoryFs(initial: Record<string, string> = {},): McpWiringFsOps & { files: Record<string, string> } {@@ -241,83 +243,86 @@ describe('checkAndRepairMcpWiringOnStartup — non-destructive decline', () => {dir = undefined;});- test('real classify on a 2^53+ integer TOML → byte-unchanged, no .broken, no-op', async () => {- dir = mkdtempSync(join(tmpdir(), 'ok-reset-repro-'));- const tomlPath = join(dir, 'config.toml');- const original = [- '# my codex config — keep my comments!',- 'model = "gpt-5"',- '',- '[mcp_servers.other]',- 'command = "node"',- 'startup_timeout_ms = 9223372036854775807',- '',- ].join('\n');- fsWriteFileSync(tomlPath, original);-- const target: EditorMcpTarget = {- id: 'codex' as McpWiringEditorId,- label: 'Codex',- format: 'toml',- topLevelKey: 'mcp_servers',- serverName: () => 'open-knowledge',- configPath: () => tomlPath,- buildEntry: () => buildManagedServerEntry({ mode: 'published' }),- scope: 'global',- };-- expect(classifyExistingMcpEntry(target, '', undefined, tomlPath)).toEqual({- kind: 'no-entry',- });-- const realFs: McpWiringFsOps = {- existsSync: (p) => fsExistsSync(p),- readFileSync: (p) => fsReadFileSync(p, 'utf8'),- writeFileSync: (p, c) => fsWriteFileSync(p, c),- mkdirSync: (p, o) => {- fsMkdirSync(p, o);- },- renameSync: (from, to) => fsRenameSync(from, to),- unlinkSync: (p) => fsUnlinkSync(p),- };- const writes: McpWiringEditorId[][] = [];- const cli: McpWiringCliSurface = {- detectInstalledEditors: () => ['codex' as McpWiringEditorId],- classifyExistingMcpEntry: () => classifyExistingMcpEntry(target, '', undefined, tomlPath),- readExistingMcpEntry: () => null,- allEditorIds: ['codex' as McpWiringEditorId],- editorTargets: { codex: target } as Record<McpWiringEditorId, EditorMcpTarget>,- writeUserMcpConfigs: async ({ editors }) => {- writes.push([...editors]);- fsWriteFileSync(tomlPath, '{"mcp_servers":{"open-knowledge":{}}}');- return editors.map((editorId) => ({- editorId,- label: editorId,- action: 'written' as const,- configPath: tomlPath,- serverName: 'open-knowledge',- }));- },- };+ test.skipIf(!NATIVE_TOML_AVAILABLE)(+ 'real classify on a 2^53+ integer TOML → byte-unchanged, no .broken, no-op',+ async () => {+ dir = mkdtempSync(join(tmpdir(), 'ok-reset-repro-'));+ const tomlPath = join(dir, 'config.toml');+ const original = [+ '# my codex config — keep my comments!',+ 'model = "gpt-5"',+ '',+ '[mcp_servers.other]',+ 'command = "node"',+ 'startup_timeout_ms = 9223372036854775807',+ '',+ ].join('\n');+ fsWriteFileSync(tomlPath, original);++ const target: EditorMcpTarget = {+ id: 'codex' as McpWiringEditorId,+ label: 'Codex',+ format: 'toml',+ topLevelKey: 'mcp_servers',+ serverName: () => 'open-knowledge',+ configPath: () => tomlPath,+ buildEntry: () => buildManagedServerEntry({ mode: 'published' }),+ scope: 'global',+ };++ expect(classifyExistingMcpEntry(target, '', undefined, tomlPath)).toEqual({+ kind: 'no-entry',+ });- const events: Array<Record<string, unknown>> = [];- const result = await checkAndRepairMcpWiringOnStartup({- isPackaged: true,- executablePath: PACKAGED_EXE,- home: dir,- platform: 'darwin',- ipcMain: inertIpcMain,- cli,- fs: realFs,- logger: { info() {}, warn() {}, error() {}, event: (e) => events.push(e) },- });+ const realFs: McpWiringFsOps = {+ existsSync: (p) => fsExistsSync(p),+ readFileSync: (p) => fsReadFileSync(p, 'utf8'),+ writeFileSync: (p, c) => fsWriteFileSync(p, c),+ mkdirSync: (p, o) => {+ fsMkdirSync(p, o);+ },+ renameSync: (from, to) => fsRenameSync(from, to),+ unlinkSync: (p) => fsUnlinkSync(p),+ };+ const writes: McpWiringEditorId[][] = [];+ const cli: McpWiringCliSurface = {+ detectInstalledEditors: () => ['codex' as McpWiringEditorId],+ classifyExistingMcpEntry: () => classifyExistingMcpEntry(target, '', undefined, tomlPath),+ readExistingMcpEntry: () => null,+ allEditorIds: ['codex' as McpWiringEditorId],+ editorTargets: { codex: target } as Record<McpWiringEditorId, EditorMcpTarget>,+ writeUserMcpConfigs: async ({ editors }) => {+ writes.push([...editors]);+ fsWriteFileSync(tomlPath, '{"mcp_servers":{"open-knowledge":{}}}');+ return editors.map((editorId) => ({+ editorId,+ label: editorId,+ action: 'written' as const,+ configPath: tomlPath,+ serverName: 'open-knowledge',+ }));+ },+ };++ const events: Array<Record<string, unknown>> = [];+ const result = await checkAndRepairMcpWiringOnStartup({+ isPackaged: true,+ executablePath: PACKAGED_EXE,+ home: dir,+ platform: 'darwin',+ ipcMain: inertIpcMain,+ cli,+ fs: realFs,+ logger: { info() {}, warn() {}, error() {}, event: (e) => events.push(e) },+ });- expect(result.status).toBe('ok');- expect(fsReadFileSync(tomlPath, 'utf8')).toBe(original);- expect(readdirSync(dir).some((name) => name.includes('.broken-'))).toBe(false);- expect(writes).toEqual([]);- expect(events.find((e) => e.event === 'mcp-config-decline')).toBeUndefined();- });+ expect(result.status).toBe('ok');+ expect(fsReadFileSync(tomlPath, 'utf8')).toBe(original);+ expect(readdirSync(dir).some((name) => name.includes('.broken-'))).toBe(false);+ expect(writes).toEqual([]);+ expect(events.find((e) => e.event === 'mcp-config-decline')).toBeUndefined();+ },+ );test('real classify on a genuinely-malformed TOML → decline, byte-unchanged, no write, bounded signal', async () => {dir = mkdtempSync(join(tmpdir(), 'ok-decline-sweep-'));diff --git a/packages/desktop/tests/integration/playwright-topass-budget.test.ts b/packages/desktop/tests/integration/playwright-topass-budget.test.tsdeleted file mode 100644index a5d4a800e..000000000--- a/packages/desktop/tests/integration/playwright-topass-budget.test.ts+++ /dev/null@@ -1,38 +0,0 @@-import { describe, expect, test } from 'bun:test';-import { spawnSync } from 'node:child_process';-import { join } from 'node:path';--const REPO_ROOT = join(__dirname, '..', '..', '..', '..');-const FIXTURE_REL = 'biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx';--describe('Invariant B — playwright-topass-budget GritQL plugin', () => {- test('fires on exactly 5 sub-15s toPass budgets (and on no negative case)', () => {- const result = spawnSync('bunx', ['biome', 'check', FIXTURE_REL], {- cwd: REPO_ROOT,- encoding: 'utf-8',- });- expect(result.status).not.toBe(0);- const output = `${result.stdout}\n${result.stderr}`;- const fires = (output.match(/Invariant B: raise the `toPass/g) ?? []).length;- expect(fires).toBe(5);- expect(output).toContain('raise the `toPass({ timeout })` budget');- expect(output).toMatch(/https?:\/\/[^\s]+/);- expect(output).toContain('biome-plugins/README.md#playwright-topass-budgetgrit');- });-- test('plugin is registered in biome.jsonc overrides, scoped to deep-link + external-link + fixture', () => {- const config = require(join(REPO_ROOT, 'biome.jsonc'));- const pluginPath = './biome-plugins/playwright-topass-budget.grit';-- const rootPlugins: string[] = config.plugins ?? [];- expect(rootPlugins).not.toContain(pluginPath);-- const overrides: Array<{ includes?: string[]; plugins?: string[] }> = config.overrides ?? [];- const matching = overrides.find((o) => (o.plugins ?? []).includes(pluginPath));- expect(matching).toBeDefined();- const includes = matching?.includes ?? [];- expect(includes).toContain('packages/desktop/tests/smoke/deep-link.e2e.ts');- expect(includes).toContain('packages/desktop/tests/smoke/external-link.e2e.ts');- expect(includes).toContain('biome-plugins/__fixtures__/playwright-topass-budget.fixture.tsx');- });-});diff --git a/packages/desktop/tests/smoke/calibration.test.ts b/packages/desktop/tests/smoke/calibration.test.tsdeleted file mode 100644index 3c55e048f..000000000--- a/packages/desktop/tests/smoke/calibration.test.ts+++ /dev/null@@ -1,91 +0,0 @@-import { describe, expect, test } from 'bun:test';-import { readdirSync } from 'node:fs';-import { resolve } from 'node:path';-import { parsePlaywrightConfigTimeout, parseTestFile } from './_helpers/parse-timeouts';--const DESKTOP_ROOT = resolve(__dirname, '..', '..');-const CONFIG_PATH = resolve(DESKTOP_ROOT, 'playwright.config.ts');-const SMOKE_DIR = resolve(DESKTOP_ROOT, 'tests', 'smoke');--const SMOKE_FILES = readdirSync(SMOKE_DIR)- .filter((f) => f.endsWith('.e2e.ts'))- .sort();--const TOPASS_BUDGET_FILES = ['deep-link.e2e.ts', 'external-link.e2e.ts'] as const;--const MIN_TOPASS_BUDGET_MS = 15_000;--describe('Playwright smoke test calibration', () => {- describe('outer per-test timeout configuration', () => {- test('playwright.config.ts has a parseable top-level timeout', () => {- const cfg = parsePlaywrightConfigTimeout(CONFIG_PATH);- expect(cfg.ci).toBeGreaterThan(0);- expect(cfg.local).toBeGreaterThan(0);- });- });-- describe('Invariant A: cumulative inner timeouts fit within outer CI budget', () => {- const cfg = parsePlaywrightConfigTimeout(CONFIG_PATH);- const outerCiMs = cfg.ci;-- for (const file of SMOKE_FILES) {- const filePath = resolve(SMOKE_DIR, file);- const fa = parseTestFile(filePath);- for (const t of fa.tests) {- test(`${file}:${t.lineNumber} ${t.testName} cumulative ≤ outer CI`, () => {- const budgetMs = t.perTestTimeoutMs ?? outerCiMs;- const breakdown = {- file,- line: t.lineNumber,- test: t.testName,- cumulativeMs: t.cumulativeMs,- budgetMs,- budgetSource:- t.perTestTimeoutMs != null ? 'test.setTimeout' : 'playwright.config.ts (CI)',- directTimeoutsMs: t.directTimeoutsMs,- helperCalls: t.helperCallNames,- tracedHelperBudgetsMs: t.tracedHelperBudgetsMs,- };- if (t.cumulativeMs > budgetMs) {- throw new Error(- `cumulative inner-timeout budget exceeds per-test budget: ${JSON.stringify(breakdown, null, 2)}`,- );- }- expect(t.cumulativeMs).toBeLessThanOrEqual(budgetMs);- });- }- }- });-- describe('Invariant B: Apple-Event / IPC toPass budgets meet minimum', () => {- for (const file of TOPASS_BUDGET_FILES) {- const filePath = resolve(SMOKE_DIR, file);- const fa = parseTestFile(filePath);- for (const t of fa.tests) {- if (t.toPassBudgetsMs.length === 0) continue;- test(`${file}:${t.lineNumber} ${t.testName} every toPass ≥ ${MIN_TOPASS_BUDGET_MS}ms`, () => {- const undersized = t.toPassBudgetsMs.filter((b) => b < MIN_TOPASS_BUDGET_MS);- if (undersized.length > 0) {- throw new Error(- `toPass({ timeout }) budget(s) below ${MIN_TOPASS_BUDGET_MS}ms minimum: ${JSON.stringify(- {- file,- line: t.lineNumber,- test: t.testName,- toPassBudgetsMs: t.toPassBudgetsMs,- undersizedMs: undersized,- minRequiredMs: MIN_TOPASS_BUDGET_MS,- },- null,- 2,- )}`,- );- }- for (const b of t.toPassBudgetsMs) {- expect(b).toBeGreaterThanOrEqual(MIN_TOPASS_BUDGET_MS);- }- });- }- }- });-});