@mmerterden/multi-agent-toolkit-mcp 3.9.0 → 3.12.0

package/CHANGELOG.md

@@ -15,6 +15,245 @@ Releases before this file exists are recorded in the git tags and commit history
 
 ---
 
+## Unreleased
+
+---
+
+## 3.12.0
+
+The CI matrix had been red for five runs. Every fix below is something the red
+leg was pointing at; three of them are defects a green local run could not see,
+because the thing that broke was the assumption that the host is a Mac.
+
+### Fixed
+
+- **`ios_app_store_audit` answered PASS on an archive it never read.** Plist
+  parsing went entirely through `plutil`, a macOS binary. Off macOS every read
+  returned `null`, all 18 rules saw an empty archive, found nothing, and the
+  verdict came back clean - on a fixture built to be non-compliant. There are
+  now three answers instead of two: `plutil` when it is there, a small XML plist
+  reader when it is not, and an explicit record of what could be read by
+  neither. An audit with a non-empty `unreadablePlists` reports
+  `measurable: false` and an `archive-readable` error, so it can no longer
+  return PASS. An unreadable archive and a compliant one now look different.
+- **Argument rules were unreachable on a host without Xcode.** `handleIOS`
+  refused everything with "Xcode not installed" before any handler ran, so the
+  checks that answer from the arguments alone - `extra_args` shell
+  metacharacters, `project` XOR `workspace`, `mode=diff` without
+  `baseline_graph`, `ios_leaks` with neither `pid` nor `bundle_id`,
+  `ios_testflight_validate` with neither `ipa_path` nor `list_providers` - never
+  ran there, and the tests that prove them could not run either. They are
+  evaluated before the capability gate now, so a malformed call is named as
+  malformed on any host. The `extra_args` character class is additionally
+  declared as `pattern` on the tool schema, where a host can enforce it too.
+- **`build-stack-plugins.mjs`-style "works on the author's machine" in the
+  test harness.** Gate 10 printed a bare exit code when a suite failed, so a red
+  CI leg said which file failed and nothing about why. It now prints the failing
+  assertions, matching both reporter shapes `node:test` uses (TAP when piped,
+  spec on a terminal).
+
+### Changed
+
+- **`validateArgs` enforces the whole schema, not three keywords of it.** It
+  checked `type`, `required` and `enum`; `pattern`, `minimum`, `maximum`,
+  `minLength`, `maxLength`, `minItems`, `maxItems` and `items` were declarable
+  and unenforced. A schema the server does not keep is worse than no schema,
+  because the host shows the caller a contract that is not real. Array `items`
+  are validated element by element, including nested `required` and property
+  types. No existing tool declared any of the newly-enforced keywords, so
+  nothing that used to be accepted is now refused.
+- **`design-check` tests are `node:test` cases.** 1002 lines and 143 assertions
+  ran inside a hand-rolled tally that ended in `process.exit(1)`: the runner saw
+  one opaque suite, coverage could not attribute a line, and a failure named the
+  file rather than the check. Same assertions, now 143 reported cases. The two
+  that need `ripgrep` register as skips with a reason instead of printing a
+  warning nobody reads.
+- **Dependency advisories closed by upgrade, not by allowlist.** `fast-uri`,
+  `hono` and `qs` reach this tree only through `@modelcontextprotocol/sdk`;
+  bounded `overrides` resolve all three. `scripts/audit-allowlist.mjs` now has
+  an empty accepted list, and its stdio-only assertion runs on every invocation
+  rather than only when an advisory happens to be open.
+- **Repo tooling only, nothing in the published package.** A `PostToolUse` hook
+  (`.claude/settings.json` -> `scripts/hooks/stdout-hygiene.sh`) greps the file
+  that was just edited for `console.log` under `index.js` and `tools/`, and
+  fails the edit back to the author when it finds one. stdout is this server's
+  JSON-RPC channel, so a stray log corrupts the protocol stream and the client
+  reports only a closed connection. Gate 5 in `scripts/gates.sh` remains the
+  authority - it decides whether the package ships; the hook only moves the
+  catch from ship-day to edit-second, and gate 5 now also asserts that the hook
+  exists, parses and is wired.
+
+---
+
+## 3.11.0
+
+### Added
+
+- **Wallet Passes, four `pass_*` tools** - `pass_validate`, `pass_build`,
+  `pass_inspect`, `pass_certificates`. Build, sign, validate and inspect Apple
+  Wallet passes; offline, deterministic, and generic by construction. All five
+  of Apple's styles are first-class with a per-style image contract, and no
+  brand, airline or project appears anywhere in the directory. 95 tools become
+  99. Three are read-only; `pass_build` writes the file the caller named.
+
+  **A literal passphrase is not an accepted input.** The caller names an
+  environment variable (`passphrase_env`) or a keychain entry, and the value
+  reaches openssl through `-passin env:` rather than as an argument, because an
+  argument is world-readable in `ps` for as long as the process runs. It is
+  never written to disk, never returned, and scrubbed out of error text. A test
+  asserts the tool surface contains no bare `passphrase` property, and another
+  asserts the value never reaches a result.
+
+  **Findings are graded by consequence, and the warning grade is the point.** An
+  error is a pass Wallet refuses. A warning is a pass Wallet accepts while
+  quietly doing less than intended, and those are the ones nobody finds: one
+  unknown `PKPassengerCapability*` value raises nothing anywhere and drops an
+  enhanced boarding pass back to the old layout; a tag renamed in iOS 26 is
+  ignored rather than rejected, so whatever it drove simply does not appear;
+  `footer@1x.png` is not the 1x footer, since Wallet looks for `footer.png`.
+
+  Run against a real signed production boarding pass, the validator reported
+  zero errors and found two genuine defects - the `@1x` file and a set of
+  labels that were half localization keys and half literal text. It also
+  reported two things wrongly on that same pass, and both were corrected rather
+  than explained: `footerFields` is a real bucket, and a dozen semantic tags it
+  called unknown are real tags.
+
+  `pass_build` refuses to build with an expired Pass Type ID certificate. Such a
+  pass builds, signs and verifies; Wallet is the first thing that refuses it,
+  and it does not say why.
+
+### Fixed
+
+Six defects found reviewing the two families above before either reached npm.
+Neither 3.10.0 nor 3.11.0 was ever published, so they are recorded here rather
+than as a patch release.
+
+- **The server pool started two servers for one root.** `withServer` read its
+  lock off a map entry that did not exist yet, so two concurrent calls for a
+  cold root each spawned a language server and the second `servers.set`
+  orphaned the first - never disposed, never swept, invisible to the idle
+  sweeper. The queue now lives in its own map, taken before any state is read.
+  `onCrash` mutated a copy, which broke identity for in-flight callers and put
+  a dead server back in the map; it mutates in place, and the crash-poison
+  window outlives the entry it belonged to.
+
+- **`dispose()` dropped an in-flight request instead of settling it.** It
+  cleared each pending timer and then the map, discarding the only reference to
+  the resolver, so the exit event that would have settled them found nothing and
+  the caller awaited forever. Only reachable against a server that ignores
+  `shutdown` - which is the case `dispose()` exists for.
+
+- **The Swift index report missed the index and guessed at its age.**
+  DerivedData folders were matched against the checkout's directory name; Xcode
+  names them after the project, and in most checkouts those differ - one
+  measured here reported `indexBuilt: false` with an 18,205-unit store on disk,
+  and printed a remedy telling the caller to build it. The candidate names now
+  come from the `.xcodeproj` / `.xcworkspace` in the root. Separately, "newest
+  unit" was sampled from the first 200 entries of a `readdir`, which is hash
+  order rather than a timeline; the directory's own mtime answers exactly, for
+  one syscall instead of 200.
+
+- **The Kotlin probe reported no JVM on a machine running Java 17.**
+  `java -version` writes to stderr and exits 0, and the version was read from
+  `execFileSync`'s return value, which is stdout. The `catch` that reads
+  `e.stderr` never ran, because nothing threw. Latent only because the probe
+  short-circuits where `kotlin-lsp` is absent - so it would have fired for
+  exactly the users who install it.
+
+- **`code_index_status` answered about paths it could not look at.** "Always
+  succeeds" is a promise about a missing toolchain, not about bad input, and it
+  had been implemented as "never validates": a relative path, a path that does
+  not exist, or literally the string `undefined` all resolved to
+  `semantic: false` with `buildSettingsSource: "fallback"` - the same answer a
+  genuinely unconfigured repo gets - and then printed a remedy telling the
+  caller to point `workspace_root` at a project root, sending them after a build
+  system that was never the problem. Every other tool in the family refused
+  those paths by name. Found by running the tools end to end rather than by
+  reading them. A call with no path at all still answers, because there is
+  nothing there to be wrong about and that is the "can this machine do anything"
+  call.
+
+- **The per-root server queue outlived the work it guarded.** The lock map
+  gained an entry for every distinct `language:root` ever asked about and
+  removed none, so a server running for days accumulated one dead promise chain
+  per root. Introduced by the fix above it, which moved the queue out of the
+  server entry precisely so it could outlive it.
+
+- **`pass_validate` filtered findings by their message text.** Checking a bare
+  `pass.json` has no files to look at, so image findings were dropped by
+  matching the sentence they were phrased with - improving the wording would
+  have silently turned them back on. They carry `kind: "asset"` now.
+
+### Changed
+
+- Gates 14c and 14d now cover `tools/pass-kit/` as well as `tools/code-intel/`:
+  no shell, and no writing to the MCP stdout channel.
+
+- **Gate 15 is new: nothing personal or corporate ships to public npm.** Every
+  tracked file is scanned for home directories, the author's username, employer
+  hostnames and day-job brands. It failed on its first run, on a design-check
+  handoff note carrying an employer app name, an internal module path, two real
+  Figma file keys and three corporate component names - a file nothing had ever
+  scanned, in a package published to the public registry.
+
+- **Gate 3b reads `README.tr.md` too, and checks the category count.** It had
+  read only `README.md`, so the Turkish page drifted three whole families
+  behind while its headline still claimed the right total, and `README.md`
+  itself said "9 categories" over ten bullets. The sum was the only number
+  anything checked.
+
+---
+
+## 3.10.0
+
+### Added
+
+- **Code Intelligence, eight `code_*` tools** - compiler-grade answers about
+  Swift and Kotlin source, over the language server each platform already
+  ships. `code_definition`, `code_references`, `code_hover`,
+  `code_document_symbols`, `code_workspace_symbols`, `code_diagnostics`,
+  `code_index_status`, `code_server_reset`. Swift needs nothing installed:
+  `sourcekit-lsp` comes with Xcode. Seven are read-only; `code_server_reset`
+  discards a cache and is marked idempotent.
+
+  One family rather than two, with the language inferred from the file
+  extension. The capability set genuinely is symmetric - both sides are LSP
+  servers answering the same methods - and duplicating the surface would double
+  the tool count for no new capability. The asymmetry lives in the output:
+  every result carries `language`, `semantic` and the resolved build-settings
+  source, and Kotlin answers carry `confidence: "alpha"`.
+
+  **What is measured, and why the tools say what they say.** On Swift 6.3.1,
+  `textDocument/references` against a two-file package answered `[]` at 0.7s
+  and `[3]` at 5.8s with nothing changed except that the background index had
+  finished. `initialize` returns `referencesProvider: true` either way, so the
+  server's own capability flag is no guide. These tools therefore wait for the
+  index and report `indexReady`, because "0 references" reads as "this symbol
+  is unused" and that is the wrong thing to be wrong about. A 658-file package
+  with dependencies had still not finished at 70s, so `index_wait_sec` is an
+  input rather than a constant. On an `.xcodeproj` root with no
+  `buildServer.json`, cross-file answers are impossible and diagnostics are
+  actively misleading; both are marked `semantic: false` with the remedy, and
+  nothing writes a build config into anyone's repository.
+
+  Kotlin runs on JetBrains `kotlin-lsp`, which is Alpha, partially closed
+  source, and not installed here or in CI. The probe and the absence path are
+  covered by tests; **the Kotlin server path is exercised by no gate in this
+  repository**, and the README says so rather than letting a green suite imply
+  coverage.
+
+- **Gate 14c** - `tools/code-intel/` must reach no shell. Gate 14b protects
+  shell templates by demanding a sanitizer on every interpolation, so it has
+  nothing to inspect in a directory that only spawns argv arrays - and would
+  have stayed silent the day one of those spawns became a `run()` template
+  taking a caller-supplied path.
+- **Gate 14d** - no code-intel module may touch `process.stdout` or inherit a
+  child's stdio. A language server speaks JSON-RPC too, and any of its stream
+  reaching fd 1 corrupts the host channel and takes all 95 tools down, not one.
+
+---
+
 ## 3.9.0
 
 ### Changed

package/README.md

@@ -7,7 +7,7 @@
 
 🇹🇷 Türkçe: [README.tr.md](./README.tr.md)
 
-**87 tools** for iOS Simulator, Android Emulator, and headless web control. MCP server that lets your AI coding assistant see, interact with, and audit your mobile apps - plus drive browsers, run an 18-rule App Store compliance audit, and orchestrate multi-step batch flows.
+**99 tools** for iOS Simulator, Android Emulator, and headless web control. MCP server that lets your AI coding assistant see, interact with, and audit your mobile apps - plus drive browsers, run an 18-rule App Store compliance audit, and orchestrate multi-step batch flows.
 
 Distributed on the **public npm registry** - `npx @mmerterden/multi-agent-toolkit-mcp` resolves with no auth, no token, no `~/.npmrc` setup.
 
@@ -15,7 +15,7 @@ Distributed on the **public npm registry** - `npx @mmerterden/multi-agent-toolki
 
 That distinction is worth keeping straight. This line once called five hosts "the multi-agent-pipeline's full-orchestration targets", which stopped being true in pipeline v10.7.0 when the Cursor / Antigravity / Codex / Copilot Chat adapters were deleted. Codex CLI returned as a first-class target in pipeline v13.0.0 on capability - it now has skills, parallel sub-agents, hooks and MCP - and it is on the verified list only because a gate exercises it, not because it is supported elsewhere. Claiming hosts nobody exercises is not free either: 2.20.0 and 2.21.0 shipped an illegal `outputSchema` on the first tool, so Claude Code rejected the whole `tools/list` and served zero of the 78 - in the primary host, for two releases, with every gate in this repo green.
 
-**7 categories:**
+**10 categories:**
 - **Device Control** (59 tools) - screenshot, tap, swipe, type, navigate, dark mode, locale, orientation, location, permissions, push notifications, real start/stop screen recording
 - **Memory** (2 tools) - `ios_leaks` runs /usr/bin/leaks against a simulator or host process, snapshot or diff against a saved memory graph; `android_meminfo` reads dumpsys meminfo, snapshot or diff. Both report `measurable:false` rather than a clean result when they could not look
 - **Crash Diagnostics** (2 tools) - `ios_list_crashes` reads the host's DiagnosticReports, `android_list_crashes` dumps the adb crash buffer, both tail-bounded
@@ -23,6 +23,8 @@ That distinction is worth keeping straight. This line once called five hosts "th
 - **Store Compliance** (5 tools) - App Store / Play Store readiness; **18-rule deep `ios_app_store_audit`** cross-references Apple ITMS error codes + App Store Review Guidelines (privacy manifest, required-reason API, Info.plist, code signing, entitlements, embedded SDK, IPv6, debug-tool leak, ...)
 - **Web Automation** (8 tools) - Playwright-powered: goto, click (CSS selectors), type, eval JS, wait for selector, extract text, screenshot. Chromium / WebKit / Firefox engines. **Requires `playwright` peer dependency.**
 - **Design Audit** (6 tools) - mock-mode vs Figma conformance: scenario inventory, mock detection, mock launch, live UI geometry, pixel/geometry/typography compare, and the HTML/PDF report with its coverage gate
+- **Code Intelligence** (8 tools) - compiler-grade answers about Swift and Kotlin source, over the language server each platform already ships: definition, references, hover, document and workspace symbols, diagnostics without a full build, plus `code_index_status` for whether this machine can answer at all. Cross-file answers need a background index, so they report whether it had settled instead of returning an empty list that reads as "unused"
+- **Wallet Passes** (4 tools) - build, sign, validate and inspect Apple Wallet passes. All five styles, offline and deterministic, and a literal passphrase is deliberately not an accepted input: the caller names an environment variable or a keychain entry. The validator grades by consequence, because the interesting failures are the silent ones - one unknown `PKPassengerCapability` value raises nothing and quietly drops an enhanced pass back to the old layout
 - **Autonomous Agent DSL** (2 tools) - `agent_run_steps` executes a batch array of {tool, args, continue_on_error?, wait_ms?} steps in one MCP round trip. Ideal for scripted login flows, form fills, multi-step QA paths. `agent_query_output` searches the full output of an earlier call that was too large to return inline, so a follow-up question does not mean re-running an expensive tool.
 
 ## Quick Start
@@ -281,6 +283,81 @@ Mock-mode vs Figma design audit. Generic and platform-agnostic - the heavy orche
 | `design_visual_compare` | Compare a Figma render vs a live screenshot without failing on size mismatch: normalizes scale, crops chrome, reports perceptual diff %, spacing/size/position (px), color (ΔE), and typography. Writes figma/live/diff/overlay/side-by-side PNGs. |
 | `design_report` | Render a design-audit report object to self-contained HTML (base64 images, CSS-positioned annotations) + optional PDF. Enforces a coverage gate: every target must be audited or skipped with an explicit reason, so a partial audit cannot render as a clean report. Wording comes from a label pack (English default, Turkish built in). |
 
+### Code Intelligence Tools (8)
+
+Compiler-grade answers about Swift and Kotlin, over the language server each
+platform already ships. Swift needs nothing installed: `sourcekit-lsp` comes
+with Xcode. Kotlin uses JetBrains `kotlin-lsp`, which is Alpha and must be
+installed separately - every Kotlin answer is labelled `confidence: "alpha"`,
+and **no gate in this repository exercises the Kotlin server**, so treat that
+side as written-but-unverified.
+
+| Tool | What it does |
+|---|---|
+| `code_definition` | Where a symbol is declared. Takes a 1-based line+column, or a `symbol` name so the caller never has to count columns. |
+| `code_references` | Every reference across the workspace. Index-backed, and reports whether the index had settled. |
+| `code_hover` | Type, signature and doc comment at a position. Works without an index for same-file and standard-library symbols. |
+| `code_document_symbols` | The symbol tree of one file. The only capability that needs no build configuration at all. |
+| `code_workspace_symbols` | Find a symbol by name across the workspace. Index-backed. |
+| `code_diagnostics` | Compiler diagnostics for one file without a full build. Says "did not look" rather than reporting 0 when nothing was published. |
+| `code_index_status` | Per language: is a server installed, which build system was found, is an index on disk, what to do if not. Always succeeds. |
+| `code_server_reset` | Stop the pooled servers. For a server wedged on stale build settings. Discards a cache, never source. |
+
+**What these can and cannot answer, by root.** sourcekit-lsp takes its compiler
+arguments from a build system, and which one it finds decides everything. The
+resolved root and how it was found come back in every result, so an answer is
+never separated from what it is worth.
+
+| Root | definition | references / workspace symbols | diagnostics |
+|---|---|---|---|
+| `Package.swift` (SwiftPM) | yes | yes, once the background index settles | yes |
+| `buildServer.json` | yes | yes | yes |
+| `.xcodeproj` with no build server | same-file only | **no** - `semantic: false` and a remedy | misleading, and marked non-semantic |
+
+Measured on Swift 6.3.1: against a two-file package, `code_references` answered
+0 at 0.7s and 3 at 5.8s, with nothing changed except that the background index
+had finished. That gap is why these tools wait for the index and report
+`indexReady` rather than returning the empty list - "0 references" reads as
+"this symbol is unused", and that is the wrong thing to be wrong about. On a
+658-file package with dependencies the first index had still not finished at
+70s, so budget accordingly with `index_wait_sec`.
+
+
+### Wallet Pass Tools (4)
+
+Build, sign, validate and inspect Apple Wallet passes. Offline, deterministic,
+and generic: all five of Apple's styles are first-class and nothing here is tied
+to a brand, an airline or a project.
+
+| Tool | What it does |
+|---|---|
+| `pass_validate` | Check a pass.json, a `.pass` directory or a built `.pkpass` against Apple's rules. Grades findings by consequence. |
+| `pass_build` | Assemble, sign (openssl, detached PKCS#7 over a SHA-1 manifest) and archive a `.pkpass`. Validates first and refuses on an error unless told otherwise. |
+| `pass_inspect` | Open a `.pkpass` and report what is in it: style, semantic tags, images and scales, localizations, manifest integrity and the signing chain. Never needs the private key. |
+| `pass_certificates` | Report the signing material: identity, expiry, and whether a passphrase source is configured. Never reads or reports key material. |
+
+**Secrets.** A literal passphrase is not an accepted parameter anywhere. The
+caller passes `passphrase_env` (the NAME of an environment variable) or
+`keychain_account`, and the value reaches openssl through `-passin env:` so it
+never appears in `ps`, never lands in a log, and is scrubbed from any error text
+on the way back.
+
+**Grading, and why it is the point.** An `error` is a pass Wallet refuses. A
+`warning` is a pass Wallet accepts while quietly doing less than intended, and
+that grade exists because those are the ones nobody finds:
+
+- one unknown `PKPassengerCapability*` value raises nothing at all and drops the
+  enhanced layout back to the old one
+- a semantic tag renamed in iOS 26 is not rejected, it is ignored, so whatever
+  it drove simply does not appear
+- `footer@1x.png` is not the 1x footer - Wallet looks for `footer.png` and never
+  finds this file, which is hashed into the manifest for nothing
+- labels that are half localization keys and half literal text localize the keys
+  and leave the rest in one language
+
+Run against a real signed production boarding pass, the validator reported zero
+errors and found both of the last two.
+
 ### Use Cases
 
 - **Visual bug hunting** - Screenshot every screen, AI analyzes for layout issues

package/README.tr.md

@@ -7,7 +7,7 @@
 
 🇬🇧 English: [README.md](./README.md)
 
-iOS Simulator, Android Emulator ve headless web kontrolü için **87 araç**. AI kodlama asistanının mobil uygulamalarını görmesini, onlarla etkileşime girmesini ve denetlemesini sağlayan bir MCP sunucusu - ayrıca tarayıcıları sürer, 18-kurallık bir App Store uyumluluk denetimi çalıştırır ve çok-adımlı batch akışlarını orkestre eder.
+iOS Simulator, Android Emulator ve headless web kontrolü için **99 araç**. AI kodlama asistanının mobil uygulamalarını görmesini, onlarla etkileşime girmesini ve denetlemesini sağlayan bir MCP sunucusu - ayrıca tarayıcıları sürer, 18-kurallık bir App Store uyumluluk denetimi çalıştırır ve çok-adımlı batch akışlarını orkestre eder.
 
 **Public npm registry** üzerinden dağıtılır - `npx @mmerterden/multi-agent-toolkit-mcp`, auth'suz, token'sız, `~/.npmrc` ayarı gerekmeden çözülür.
 
@@ -15,13 +15,16 @@ iOS Simulator, Android Emulator ve headless web kontrolü için **87 araç**. AI
 
 Bu ayrımı net tutmakta fayda var. Bu satır bir zamanlar beş host'u "multi-agent-pipeline'ın tam-orkestrasyon hedefleri" olarak adlandırıyordu, bu da pipeline v10.7.0'da Cursor / Antigravity / Codex / Copilot Chat adaptörleri silindiğinde doğru olmaktan çıktı. Codex CLI, pipeline v13.0.0'da yetenek üzerinden birinci-sınıf bir hedef olarak geri döndü - artık skill'leri, paralel sub-agent'ları, hook'ları ve MCP'si var - ve doğrulanmış listede sadece bir kapı onu çalıştırdığı için yer alıyor, başka bir yerde desteklendiği için değil. Kimsenin çalıştırmadığı host'ları iddia etmek de bedavaya gelmiyor: 2.20.0 ve 2.21.0, ilk araçta yasa dışı bir `outputSchema` gönderdi, bu yüzden Claude Code tüm `tools/list`'i reddetti ve 78'in sıfırını sundu - birincil host'ta, iki release boyunca, bu repo'daki her kapı yeşilken.
 
-**7 kategori:**
+**10 kategori:**
 - **Device Control** (59 araç) - screenshot, tap, swipe, type, navigate, dark mode, locale, orientation, location, permissions, push notifications, gerçek start/stop ekran kaydı
+- **Memory** (2 araç) - `ios_leaks`, bir simülatör ya da host süreci üzerinde /usr/bin/leaks çalıştırır; anlık görüntü alır ya da kayıtlı bir memory graph ile karşılaştırır. `android_meminfo`, dumpsys meminfo okur, aynı şekilde anlık görüntü ya da fark. İkisi de bakamadıklarında temiz bir sonuç değil `measurable:false` döner
 - **Crash Diagnostics** (2 araç) - `ios_list_crashes` host'un DiagnosticReports dizinini okur, `android_list_crashes` adb crash buffer'ını döker, ikisi de tail-sınırlı
-- **Accessibility Audit** (2 araç) - eksik label'ları, küçük tap target'larını, eksik identifier'ları otomatik tespit et (scope filtreleme desteklenir)
+- **Accessibility Audit** (3 araç) - `ios_accessibility_audit` / `android_accessibility_audit` canlı ağacı okur: eksik label, ekran okuyucunun adlandıramadığı kontrol, küçük tap target, eksik identifier ve görsel düzeni izlemeyen okuma sırası. `ios_accessibility_audit_deep` ise ağacın gösteremeyeceği şeyler için Apple'ın kendi XCUIAccessibilityAudit'ini XCUITest'iniz üzerinden çalıştırır: kontrast, Dynamic Type, kırpılmış metin. Üçü de bakamadıklarında temiz bir sonuç değil, gerekçesiyle `measurable:false` döner
 - **Store Compliance** (5 araç) - App Store / Play Store hazırlığı; **18-kurallık derin `ios_app_store_audit`**, Apple ITMS hata kodları + App Store Review Guidelines'a çapraz referans verir (privacy manifest, required-reason API, Info.plist, code signing, entitlements, gömülü SDK, IPv6, debug-tool sızıntısı, ...)
 - **Web Automation** (8 araç) - Playwright-destekli: goto, click (CSS selector'lar), type, JS eval, selector bekleme, metin çıkarma, screenshot. Chromium / WebKit / Firefox motorları. **`playwright` peer dependency'si gerektirir.**
 - **Design Audit** (6 araç) - mock-mode vs Figma uygunluğu: scenario envanteri, mock tespiti, mock launch, canlı UI geometrisi, piksel/geometri/tipografi karşılaştırması, ve coverage kapısıyla birlikte HTML/PDF rapor
+- **Code Intelligence** (8 araç) - Swift ve Kotlin kaynağı hakkında derleyici seviyesinde cevaplar, her platformun zaten getirdiği dil sunucusu üzerinden: tanım, referanslar, hover, dosya ve proje geneli semboller, tam build olmadan tanılamalar, ve bu makinenin soruyu cevaplayıp cevaplayamayacağını söyleyen `code_index_status`. Çapraz dosya cevapları arka plan index'ine bağlı olduğu için, "kullanılmıyor" diye okunan boş bir liste yerine index'in oturup oturmadığını raporlar
+- **Wallet Passes** (4 araç) - Apple Wallet pass'i üret, imzala, doğrula ve incele. Beş stilin hepsi, çevrimdışı ve deterministik; düz metin parola bilerek kabul edilen bir girdi değil: çağıran taraf bir ortam değişkeninin adını ya da bir keychain kaydını verir. Doğrulayıcı bulguları sonucuna göre derecelendirir, çünkü asıl önemli hatalar sessiz olanlar - tanınmayan tek bir `PKPassengerCapability` değeri hiçbir uyarı üretmez ve gelişmiş pass'i sessizce eski düzene düşürür
 - **Autonomous Agent DSL** (2 araç) - `agent_run_steps`, tek bir MCP round trip'inde {tool, args, continue_on_error?, wait_ms?} adımlarından oluşan bir batch dizisini çalıştırır. Scriptlenmiş login akışları, form doldurma, çok-adımlı QA yolları için ideal. `agent_query_output`, satır içi dönemeyecek kadar büyük olan önceki bir çağrının tam çıktısını arar; böylece bir takip sorusu pahalı aracı yeniden koşturmak anlamına gelmez.
 
 ## Hızlı Başlangıç