pacphi
diff --git a/‎docs/TROUBLESHOOTING.md‎
Lines changed: 1 addition & 0 deletions b/‎docs/TROUBLESHOOTING.md‎
Lines changed: 1 addition & 0 deletions
diff --git a/‎src/commands/status/sections/hosts.mjs‎
Lines changed: 31 additions & 21 deletions b/‎src/commands/status/sections/hosts.mjs‎
Lines changed: 31 additions & 21 deletions
diff --git a/‎src/commands/sync.mjs‎
Lines changed: 28 additions & 15 deletions b/‎src/commands/sync.mjs‎
Lines changed: 28 additions & 15 deletions
diff --git a/‎src/lib/aqe-embedding-toml.mjs‎
Lines changed: 7 additions & 7 deletions b/‎src/lib/aqe-embedding-toml.mjs‎
Lines changed: 7 additions & 7 deletions
diff --git a/‎src/lib/codex-toml-safety.mjs‎
Lines changed: 27 additions & 0 deletions b/‎src/lib/codex-toml-safety.mjs‎
Lines changed: 27 additions & 0 deletions
diff --git a/‎src/lib/heal.mjs‎
Lines changed: 11 additions & 4 deletions b/‎src/lib/heal.mjs‎
Lines changed: 11 additions & 4 deletions
diff --git a/‎src/lib/mcp.mjs‎
Lines changed: 8 additions & 13 deletions b/‎src/lib/mcp.mjs‎
Lines changed: 8 additions & 13 deletions
@@ -36,6 +36,7 @@ ak sync             # apply it
 | Codex receives automatic deja-vu recall while Agentic Kit says MCP mode | A user-owned Codex deja-vu plugin can contribute session/per-prompt/precompaction hooks independently of Agentic Kit's mode | disable/remove that plugin through Codex if MCP-only behavior is required. `ak sync` preserves external plugins and reports the effective auto surface without claiming a fix |
 | `--purge-deja-vu-data` refuses the index path | The observed path is broad, relative, outside an approved data root, overlaps config/transcript sources, or crosses a symlink | move/reconfigure the derived index safely, run `deja doctor --offline`, then retry. Never bypass the guard by deleting a host transcript root |
 | Just upgraded ruflo/agentic-qe (`npm i -g …`) and things feel off | Upgrades re-resolve dependencies: native SQLite bindings and the aidefence package get dropped, and ruflo's helper auto-refresh regenerates the statusline without the footer | `ak sync` (this is its main job) |
+| `status` shows a host `installed but not executable` | npm exits 0 even when an optional dependency fails, so a package can be recorded without its platform binary. Codex ships its binary as per-platform versions (for example `@openai/codex-darwin-arm64`) published minutes after the main version, so an upgrade in that window can leave `codex` unable to start | `ak sync` reinstalls an npm-owned host and verifies it starts; upgrades and installs already retry once with `--prefer-online`. An external (mise/native/brew) install is reinstalled with its own tool |
 | `status` shows `natives … WASM fallback` | agentdb resolved a non-native better-sqlite3 — on this path **memory writes can silently vanish**. Common causes are npm ≥11.17 blocking install scripts during upgrades, or a stale better-sqlite3 ≤12.9 pin on Node 26 | `ak sync` selects a Node-compatible release and installs the native binding |
 | `status` shows `aidefence missing` | ruflo ≥3.28 stopped shipping `@claude-flow/aidefence` but `ruflo security defend` still imports it — injection defense is silently non-functional ([ruvnet/ruflo#2670](https://github.com/ruvnet/ruflo/issues/2670)) | `ak sync` reinstalls it; `ak x verify security` proves defend works (exit 1=threat / 0=clean) |
 | `status` shows oversized RVF store(s) | A runaway append after a hard exit grew a `.rvf` past the 2 GB cap (seen at ~277 GB once) | `ak sync` quarantines the oversized store; agentic-qe rebuilds it |
 
@@ -1,38 +1,48 @@
 // hosts (install-if-missing) — cheap: file read + `which`, no network.
 // An enabled host that is entirely absent is installable by sync; an external
-// install (mise/native/brew) is reported but never touched.
-import { HOSTS, hostInstallState, hostAuthState } from '../../../lib/providers.mjs';
+// install (mise/native/brew) is reported but never touched. An npm install is
+// also launched once (`--version`): its package.json can outlive the binary.
+import { HOSTS, hostInstallState, hostAuthState, hostExecutable } from '../../../lib/providers.mjs';
 import { row } from '../row.mjs';
 
+const DEFAULT_DEPS = { installState: hostInstallState, executable: hostExecutable, authState: hostAuthState };
+
+// Install row + auth row for a host that is on disk.
+async function installedHostRows(h, st, primary, deps) {
+  const label = `${h.id} ${st.version ?? ''} (${st.method}${st.method === 'external' ? ' — self-managed' : ''})`;
+  const launch = st.method === 'npm' ? await deps.executable(h) : { ok: true, detail: null };
+  const install = launch.ok ? row('hosts', 'ok', label)
+    : row('hosts', primary ? 'fail' : 'warn', `${label} installed but not executable: ${launch.detail}`,
+      `sync reinstalls ${h.pkg}`);
+  // auth mode (billing axis): oauth/subscription ($0) vs metered api-key.
+  // A distinct row so `ak status --json` (and the dashboard) can badge it.
+  const auth = deps.authState(h.id, { present: true });
+  const billing = auth.billing === 'subscription' ? 'subscription, $0'
+    : auth.billing === 'metered' ? 'metered' : auth.billing;
+  return [install, row('hosts', auth.mode === 'none' ? 'warn' : 'ok',
+    `${h.id} auth: ${auth.mode} (${billing})${auth.source ? ` · ${auth.source}` : ''}${auth.note ? ` — ${auth.note}` : ''}`,
+    auth.mode === 'none' ? `${h.id} login` : null)];
+}
+
 export default {
   id: 'hosts',
-  async collect({ cfg, integrationFacts }) {
+  /** @param {{ cfg: any, integrationFacts: any, hostDeps?: Partial<typeof DEFAULT_DEPS> }} ctx */
+  async collect({ cfg, integrationFacts, hostDeps = {} }) {
+    const deps = { ...DEFAULT_DEPS, ...hostDeps };
     const rows = [];
     try {
       // primary host absent = fail (nothing can drive); alternate absent = warn.
       const primaryHost = cfg.routing?.primaryHost ?? 'claude';
       for (const h of HOSTS) {
         if (!cfg.integrations.hosts[h.id]) continue;
-        const detected = integrationFacts.hosts[h.id];
-        if (detected?.present === false) {
-          rows.push(row('hosts', h.id === primaryHost ? 'fail' : 'warn',
-            `${h.id} enabled but not installed${h.id === primaryHost ? ' (primary)' : ''}`, `sync installs ${h.pkg}`));
-          continue;
-        }
-        const st = await hostInstallState(h);
+        const primary = h.id === primaryHost;
+        const st = integrationFacts.hosts[h.id]?.present === false
+          ? { method: 'absent', version: null } : await deps.installState(h);
         if (st.method === 'absent') {
-          rows.push(row('hosts', h.id === primaryHost ? 'fail' : 'warn',
-            `${h.id} enabled but not installed${h.id === primaryHost ? ' (primary)' : ''}`, `sync installs ${h.pkg}`));
+          rows.push(row('hosts', primary ? 'fail' : 'warn',
+            `${h.id} enabled but not installed${primary ? ' (primary)' : ''}`, `sync installs ${h.pkg}`));
         } else {
-          rows.push(row('hosts', 'ok', `${h.id} ${st.version ?? ''} (${st.method}${st.method === 'external' ? ' — self-managed' : ''})`));
-          // auth mode (billing axis): oauth/subscription ($0) vs metered api-key.
-          // A distinct row so `ak status --json` (and the dashboard) can badge it.
-          const auth = hostAuthState(h.id, { present: true });
-          const billing = auth.billing === 'subscription' ? 'subscription, $0'
-            : auth.billing === 'metered' ? 'metered' : auth.billing;
-          rows.push(row('hosts', auth.mode === 'none' ? 'warn' : 'ok',
-            `${h.id} auth: ${auth.mode} (${billing})${auth.source ? ` · ${auth.source}` : ''}${auth.note ? ` — ${auth.note}` : ''}`,
-            auth.mode === 'none' ? `${h.id} login` : null));
+          rows.push(...await installedHostRows(h, st, primary, deps));
         }
       }
     } catch (e) {
 
@@ -19,7 +19,7 @@ import { companionLifecycleFor } from '../lib/adapters/companion-lifecycle-regis
 import { renderApplyReport } from '../lib/adapters/lifecycle-render.mjs';
 import { listDaemons, staleDaemons, reap } from '../lib/daemons.mjs';
 import { loadKitConfig, saveKitConfig } from '../lib/config.mjs';
-import { commandHosts, hostInstallState, installHost, convergeProviderStack, guidanceContext, reportRetiredRouteChanges } from '../lib/providers.mjs';
+import { HOSTS, commandHosts, hostInstallState, hostExecutable, installHost, convergeProviderStack, guidanceContext, reportRetiredRouteChanges } from '../lib/providers.mjs';
 import { driftReport, selfDrift } from '../lib/versions.mjs';
 import { drift as ruvnetBrainDrift } from '../lib/ruvnet-brain.mjs';
 import { RUVECTOR_PKG, managed as ruvectorManaged } from '../lib/ruvector.mjs';
@@ -126,6 +126,14 @@ Examples:
 // report, step, state}. `state` carries the two cross-step signals
 // (`dejaVuApplyFailed`, `aqeRouterApplyFailure`) the final convergence check
 // needs — the only state that survives past its own step.
+const HOST_LIFECYCLE = { installState: hostInstallState, executable: hostExecutable, install: installHost };
+
+/** A host package's CLI must start after an upgrade, not just extract. */
+export function hostUpgradeOptions(pkg) {
+  const host = HOSTS.find((h) => h.pkg === pkg);
+  return host ? { bin: host.bin } : {};
+}
+
 export const SYNC_STEPS = [
   {
     id: 'agent-browser',
@@ -141,6 +149,24 @@ export const SYNC_STEPS = [
       return result;
     },
   },
+  // hosts: install any ENABLED host that is entirely absent, and reinstall an
+  // npm-owned host whose CLI cannot start (npm exits 0 after dropping a failed
+  // optional platform binary). External installs are never touched; updates
+  // ride the `versions` step via driftReport. Runs before the Codex MCP and
+  // provider steps, which shell out to the host CLIs.
+  {
+    id: 'hosts',
+    when: (subs) => subs.has('hosts'),
+    run: async (ctx) => {
+      const { installState, executable, install } = { ...HOST_LIFECYCLE, ...ctx.hostLifecycle };
+      for (const h of commandHosts()) {
+        if (!ctx.cfg.integrations.hosts[h.id]) continue;
+        const { method } = await installState(h);
+        if (method === 'absent') await ctx.step(`install ${h.id}`, () => install(h.id));
+        else if (method === 'npm' && !(await executable(h)).ok) await ctx.step(`repair ${h.id}`, () => install(h.id));
+      }
+    },
+  },
   {
     id: 'codex-mcp-repair',
     when: (subs) => subs.has('codex-mcp'),
@@ -170,7 +196,7 @@ export const SYNC_STEPS = [
       // No force here: the pre-plan refresh in run() already ran for every
       // non-dry-run, non-no-upgrade sync, so this read hits that fresh cache.
       for (const d of await driftReport()) {
-        if (d.outdated || !d.installed) await ctx.step(`upgrade ${d.pkg}`, () => heal.upgradePackage(d.pkg));
+        if (d.outdated || !d.installed) await ctx.step(`upgrade ${d.pkg}`, () => heal.upgradePackage(d.pkg, hostUpgradeOptions(d.pkg)));
       }
     },
   },
@@ -283,19 +309,6 @@ export const SYNC_STEPS = [
       }
     },
   },
-  // hosts: install any ENABLED host that is entirely absent (updates to
-  // npm-managed hosts ride the `versions` step above via driftReport).
-  {
-    id: 'hosts',
-    when: (subs) => subs.has('hosts'),
-    run: async (ctx) => {
-      for (const h of commandHosts()) {
-        if (!ctx.cfg.integrations.hosts[h.id]) continue;
-        if ((await hostInstallState(h)).method !== 'absent') continue;
-        await ctx.step(`install ${h.id}`, () => installHost(h.id));
-      }
-    },
-  },
   // Managed companion convergence is independent from host lifecycle
   // adapters. The adapter owns exact package/target/index ordering and mutates
   // only its in-memory ownership ledger; this command owns persistence. Save a
 
@@ -1,5 +1,5 @@
 // Deliberately narrow TOML editor: unsupported encodings remain user-owned.
-import { inspectCodexTomlStructure, isTomlTableLine } from './codex-toml-safety.mjs';
+import { inspectCodexTomlStructure, isTomlTableLine, tomlStringArrayAt } from './codex-toml-safety.mjs';
 import { recognizedAqeTransport, parseEmbeddingJson } from './aqe-embedding-transport.mjs';
 const BASE = 'mcp_servers.agentic-qe';
 const ENV = `${BASE}.env`;
@@ -32,18 +32,18 @@ function scalar(text, key) {
   return parseEmbeddingJson(match[1]);
 }
 
-function transportAssignment(text, transport) {
+function transportAssignment(text, transport, rest) {
   if (/^env\s*=/.test(text)) throw new Error('inline AQE environment requires manual embedding configuration');
   if (/^command\s*=/.test(text)) {
     if (transport.command !== null) throw new Error('duplicate AQE command');
     transport.command = scalar(text, 'command');
   }
   if (/^args\s*=/.test(text)) {
     if (transport.args !== null) throw new Error('duplicate AQE arguments');
-    const match = /^args\s*=\s*(\[[^\n]*\])\s*(?:#.*)?$/.exec(text);
-    if (!match) throw new Error('unsupported AQE arguments encoding');
-    transport.args = parseEmbeddingJson(match[1]);
-    if (!Array.isArray(transport.args)) throw new Error('unsupported AQE arguments shape');
+    // `rest` starts at this line so a multi-line array is read whole.
+    const args = tomlStringArrayAt(rest, 'args');
+    if (!args?.value) throw new Error('unsupported AQE arguments encoding');
+    transport.args = args.value;
   }
 }
 
@@ -73,7 +73,7 @@ export function aqeTomlEnvironment(source) {
     if (table === 'unrelated') continue;
     // Dotted/quoted keys can alias a managed table: refuse rather than guessing.
     if (!/^[A-Za-z0-9_-]+\s*=/.test(text)) throw new Error('dotted or quoted TOML assignments require manual embedding configuration');
-    if (table === BASE) transportAssignment(text, transport);
+    if (table === BASE) transportAssignment(text, transport, source.slice(line.start));
     if (table === ENV && new RegExp(`^${KEY}\\s*=`).test(text)) {
       if (endpoint) throw new Error('duplicate AQE endpoint');
       endpoint = { ...line, value: scalar(text, KEY) };
 
@@ -197,3 +197,30 @@ export function inspectCodexTomlStructure(source) {
 }
 
 export const isTomlTableLine = (line) => TABLE.test(line) || ARRAY_TABLE.test(line);
+
+// Arrays of basic strings, single- or multi-line (toml_edit, which Codex and
+// AQE write through, puts each element on its own line with a trailing comma).
+// Comments and non-string elements inside the array stay unsupported.
+const ARRAY_STRING = '"(?:[^"\\\\\\r\\n]|\\\\.)*"';
+const STRING_ARRAY = `\\[\\s*(?:${ARRAY_STRING}(?:\\s*,\\s*${ARRAY_STRING})*\\s*,?\\s*)?\\]`;
+
+function parseStringArray(value) {
+  try {
+    const parsed = JSON.parse(value.replace(/,\s*\]$/, ']'));
+    return Array.isArray(parsed) && parsed.every((entry) => typeof entry === 'string') ? parsed : null;
+  } catch { return null; }
+}
+
+/** Read `key = [..strings..]` at the very start of `text`. Returns
+ * { value, text } where value is null for an unsupported encoding, or null when
+ * `text` does not start with that assignment. */
+export function tomlStringArrayAt(text, key) {
+  const match = new RegExp(`^[\\t ]*${key}[\\t ]*=[\\t ]*(${STRING_ARRAY})[\\t ]*(?:#[^\\r\\n]*)?(?=\\r?\\n|$)`).exec(text);
+  return match ? { value: parseStringArray(match[1]), text: match[0] } : null;
+}
+
+/** Find the first `key = [..strings..]` line in a table body (see tomlStringArrayAt). */
+export function findTomlStringArray(body, key) {
+  const match = new RegExp(`^[\\t ]*${key}[\\t ]*=[\\t ]*(${STRING_ARRAY})[\\t ]*(?:#[^\\r\\n]*)?$`, 'm').exec(body);
+  return match ? { value: parseStringArray(match[1]), text: match[0] } : null;
+}
@@ -14,7 +14,7 @@ import { KIT_PKG } from './versions.mjs';
 import { scanRvf, quarantine } from './rvf.mjs';
 import { INSTALL_SPEC, INSTALL_ARGS, RELEASE_ASSET as RB_RELEASE_ASSET, NIGHTLY_LABEL as RB_NIGHTLY_LABEL, nightlyAgentPlist as rbNightlyPlist, present as rbPresent, latestRelease as rbLatestRelease, recordInstalledRelease as rbRecord } from './ruvnet-brain.mjs';
 import { PKG as ADB_PKG, present as adbPresent, coherence as adbCoherence } from './agentdb.mjs';
-import { globalInstallArgs } from './npm-global-install.mjs';
+import { globalInstallArgs, installGlobalCli } from './npm-global-install.mjs';
 
 // NB: `--allow-scripts` is rejected for project-scoped installs (EALLOWSCRIPTS,
 // npm >=11.17) — it is a global-install flag only. Plain installs still get
@@ -153,9 +153,16 @@ export function healRvf(projectAqeDir) {
   return { ok: true, detail: removed.length ? `quarantined: ${removed.join(', ')}` : 'healthy' };
 }
 
-/** Upgrade a global package to latest (with allow-scripts). */
-export async function upgradePackage(pkg) {
-  const r = await run('npm', globalInstallArgs(`${pkg}@latest`),
+/** Upgrade a global package to latest (with allow-scripts). With `bin`, the
+ *  package's CLI must also start afterwards (see installGlobalCli).
+ *  @param {string} pkg
+ *  @param {{ bin?: string|null, runner?: typeof run, sleep?: (ms: number) => Promise<void> }} [opts] */
+export async function upgradePackage(pkg, { bin = null, runner = run, sleep } = {}) {
+  if (bin) {
+    const r = await installGlobalCli(`${pkg}@latest`, bin, { runner, sleep });
+    return { ok: r.ok, detail: r.ok ? (r.retried ? 'upgraded (missing platform files repaired on retry)' : 'upgraded') : r.detail };
+  }
+  const r = await runner('npm', globalInstallArgs(`${pkg}@latest`),
     { timeout: 600_000 });
   return { ok: r.code === 0, detail: r.code === 0 ? 'upgraded' : r.stderr.split('\n').slice(-3).join(' ') };
 }
 
@@ -13,6 +13,7 @@ import { writeFileWithBackup } from './file-write.mjs';
 import { managedAgentBrowserEnv } from './agent-browser.mjs';
 import { isRufloMcpTransport } from './ruflo-mcp-transport.mjs';
 import { retiredCodexTransport } from './host-alignment.mjs';
+import { findTomlStringArray } from './codex-toml-safety.mjs';
 
 /** Enumerate MCP tool names from the installed package's mcp-tools modules,
  *  grouped by name prefix (family). Returns Map<family, string[]>. */
@@ -166,14 +167,6 @@ function tomlString(value) {
   try { return JSON.parse(value); } catch { return null; }
 }
 
-function tomlStringArray(value) {
-  if (!value) return null;
-  try {
-    const parsed = JSON.parse(value);
-    return Array.isArray(parsed) && parsed.every((entry) => typeof entry === 'string') ? parsed : null;
-  } catch { return null; }
-}
-
 const sameArgs = (left, right) => JSON.stringify(left) === JSON.stringify(right);
 const fingerprint = (value) => createHash('sha256').update(value).digest('hex');
 
@@ -184,7 +177,7 @@ function mcpTableName(table) {
 
 /** Read the bounded base-table sections behind Codex MCP registrations. This
  * is deliberately not a general TOML parser: only a base
- * `[mcp_servers.<name>]` table with single-line string command/args facts is
+ * `[mcp_servers.<name>]` table with string command and string-array args facts is
  * observed. Extra fields or child tables preserve ownership, except for the
  * exact kit-managed browser environment on the retired Ruflo transport. */
 function codexMcpSections(file, scope) {
@@ -203,10 +196,13 @@ function codexMcpSections(file, scope) {
     const bodyEnd = headers[index + 1]?.index ?? source.length;
     const body = source.slice(bodyStart, bodyEnd);
     const command = tomlString(/^\s*command\s*=\s*("(?:[^"\\]|\\.)*")\s*$/m.exec(body)?.[1]);
-    const args = tomlStringArray(/^\s*args\s*=\s*(\[[^\n]*\])\s*$/m.exec(body)?.[1]);
+    const argsAssignment = findTomlStringArray(body, 'args');
+    const args = argsAssignment?.value ?? null;
     const enabledValue = /^\s*enabled\s*=\s*(true|false)\s*(?:#.*)?$/m.exec(body)?.[1];
     const enabled = enabledValue == null ? undefined : enabledValue === 'true';
-    const meaningful = body.split(/\r?\n/).map((line) => line.trim())
+    // Count a multi-line args array as the single field it is.
+    const fieldBody = argsAssignment ? body.replace(argsAssignment.text, 'args = []') : body;
+    const meaningful = fieldBody.split(/\r?\n/).map((line) => line.trim())
       .filter((line) => line && !line.startsWith('#'));
     const exactFields = meaningful.length === 2
       && meaningful.some((line) => /^command\s*=/.test(line))
@@ -445,9 +441,8 @@ export function rufloCodexMcpStatus(cfg, { home = os.homedir() } = {}) {
       const next = rest.search(/^\s*\[/m);
       const body = rest.slice(0, next < 0 ? rest.length : next);
       const commandMatch = /^\s*command\s*=\s*("(?:[^"\\]|\\.)*")\s*$/m.exec(body);
-      const argsMatch = /^\s*args\s*=\s*(\[[^\n]*\])\s*$/m.exec(body);
       try { if (commandMatch) command = JSON.parse(commandMatch[1]); } catch { /* non-canonical TOML */ }
-      try { if (argsMatch) args = JSON.parse(argsMatch[1]); } catch { /* non-canonical TOML */ }
+      args = findTomlStringArray(body, 'args')?.value ?? null;
     }
   } catch { /* config absent → not registered */ }
   return {