@rubytech/create-maxy-lite 0.1.14 → 0.1.16

package/lib/orchestrate.mjs

@@ -15,9 +15,9 @@
 // All external effects go through injected ctx primitives so the sequence,
 // guards and idempotency are unit-testable; index.mjs wires the real ones.
 
-import { PATHS, launcherPath, ptyLoadCommand, GUEST_BUILD_ENV } from './paths.mjs'
+import { PATHS, ptyLoadCommand, GUEST_BUILD_ENV } from './paths.mjs'
 
-export const STEP_NAMES = ['termux-deps', 'proot', 'ubuntu', 'node', 'toolchain', 'vault-bind', 'npm-app']
+export const STEP_NAMES = ['termux-deps', 'proot', 'ubuntu', 'node', 'toolchain', 'vault-bind', 'app-code', 'npm-app']
 
 /** Throw unless the command result is a clean exit; the message carries the command's stderr. */
 function ensureOk(r) {
@@ -111,20 +111,32 @@ const STEPS = [
       ensureOk(c.run(`proot-distro login ${PATHS.distro} --bind ${PATHS.vaultHost}:${PATHS.vaultGuest} -- test -d ${PATHS.vaultGuest}`))
     },
   },
+  {
+    // The app code — payload (server.mjs, validator, schema, skills), the skills
+    // symlinks, and the launcher — is cheap to (re)lay and must land on EVERY run,
+    // including a re-install of an already-converged device. It lived inside the
+    // guarded npm-app step, so once node-pty/claude/launcher were present npm-app
+    // skipped and a changed server.mjs never reached the device (a new build had
+    // "no effect"). Always-run (`done: () => false`); all three ops overwrite
+    // idempotently. Ordered before npm-app because installDeps builds in the dir
+    // this lays.
+    name: 'app-code',
+    done: () => false,
+    run: async (c) => {
+      c.layPayload()
+      c.linkSkills()
+      c.writeLauncher()
+    },
+  },
   {
     name: 'npm-app',
-    // Converged only when node-pty's native module LOADS, claude is present, AND
-    // the launcher is written. The deps probe is `require("node-pty")`, not a
-    // file-presence test: a present-but-unloadable `pty.node` (stale/wrong-ABI
-    // from a prior build) would pass `test -f` yet crash the relay at launch, so
-    // the guard would skip the very install that repairs it. The load-check is the
-    // same condition `installDeps` asserts post-build (shared ptyLoadCommand), so
-    // the guard admits exactly what the build proves; a non-loadable module fails
-    // the guard and the step re-installs.
-    done: (c) =>
-      c.runIn(ptyLoadCommand()).code === 0 &&
-      c.runIn('command -v claude').code === 0 &&
-      c.existsHost(launcherPath()),
+    // The heavy, network-bound installs only. Converged when node-pty's native
+    // module LOADS and claude is present (the load-check is `require("node-pty")`,
+    // the same condition installDeps asserts post-build, so a present-but-
+    // unloadable pty.node fails the guard and the step rebuilds). The launcher +
+    // payload are handled by the always-run app-code step, so they are not part of
+    // this guard.
+    done: (c) => c.runIn(ptyLoadCommand()).code === 0 && c.runIn('command -v claude').code === 0,
     run: async (c) => {
       ensureOk(await c.runInStream(`npm install -g @anthropic-ai/claude-code@${c.pins.claudeCode}`))
       // Pin ttyd to the manifest version via its release binary; apt would float.
@@ -133,12 +145,7 @@ const STEPS = [
           `curl -fsSL -o /usr/local/bin/ttyd https://github.com/tsl0922/ttyd/releases/download/${c.pins.ttyd}/ttyd.aarch64 && chmod +x /usr/local/bin/ttyd`,
         ),
       )
-      c.layPayload()
-      // The payload now carries the skills tree into the app dir; link each skill
-      // into the on-device personal-skills source so `claude` discovers them.
-      c.linkSkills()
       await c.installDeps()
-      c.writeLauncher()
     },
   },
 ]

package/lib/paths.mjs

@@ -183,7 +183,14 @@ export function launcherScript({
   port = PATHS.webchatPort,
   distro = PATHS.distro,
 } = {}) {
-  const inner = `cd ${webchatDir} && ${webchatRunEnv({ vaultGuest, port })} exec node server.mjs`
+  // Kill any prior relay before starting so `maxy-lite` always runs the freshly
+  // installed server.mjs, never a stale process still holding the port (which made
+  // a new build appear to "have no effect"). The relay writes its pid to
+  // <vault>/.maxy-lite/relay.pid; kill via the bash builtin (no external tools),
+  // and a best-effort pkill catches a relay started before pid-file support. The
+  // relay's own EADDRINUSE handler makes any remaining conflict loud.
+  const killPrior = `P=$(cat ${vaultGuest}/.maxy-lite/relay.pid 2>/dev/null); [ -n "$P" ] && kill "$P" 2>/dev/null; pkill -f "node server.mjs" 2>/dev/null; sleep 1`
+  const inner = `${killPrior}; cd ${webchatDir} && ${webchatRunEnv({ vaultGuest, port })} exec node server.mjs`
   return `#!${prefix}/bin/bash
 # maxy-lite launcher — starts the on-device web chat relay.
 # Enters the glibc Ubuntu layer with the shared-storage vault bound in, then runs

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rubytech/create-maxy-lite",
-  "version": "0.1.14",
+  "version": "0.1.16",
   "description": "Install maxy-lite on an Android phone: orchestrates proot-distro Ubuntu, glibc Node, claude, the web-chat relay, the vault and its bind-mount — run via npx in bare Termux.",
   "type": "module",
   "engines": {

package/payload/webchat/server.mjs

@@ -34,6 +34,7 @@ const CLAUDE_BIN = process.env.LITE_CLAUDE_BIN || 'claude'
 const PROJECTS_DIR = path.join(os.homedir(), '.claude', 'projects')
 const STATE_DIR = process.env.LITE_STATE_DIR || path.join(AGENT_HOME, '.maxy-lite')
 const SESSION_FILE = path.join(STATE_DIR, 'session-id')
+const PID_FILE = path.join(STATE_DIR, 'relay.pid') // so the launcher can kill a prior relay before starting a fresh one
 const TAIL_MS = 250 // JSONL poll interval
 const QUIET_MS = 1500 // silence that ends a turn for op=done
 const NO_REPLY_MS = 8000 // a turn with NO transcript growth this long = swallowed; surface why
@@ -261,6 +262,22 @@ wss.on('connection', (ws) => {
   })
 })
 
+// A stale relay still holding the port is the usual cause of "new build installed
+// but old code serving". Make it loud instead of a silent stale-serve.
+server.on('error', (e) => {
+  log('listen-error', { code: e.code, port: PORT })
+  process.exit(1)
+})
 server.listen(PORT, '127.0.0.1', () => {
+  // Record our pid so the next launch can kill this relay and take over cleanly,
+  // and emit a build marker: op=relay-start proves THIS server.mjs (not a stale
+  // one) is the live process.
+  try {
+    fs.mkdirSync(STATE_DIR, { recursive: true })
+    fs.writeFileSync(PID_FILE, String(process.pid))
+  } catch {
+    /* pid file is best-effort */
+  }
+  log('relay-start', { pid: process.pid, port: PORT })
   log('listening', { port: PORT, sessionId, home: AGENT_HOME })
 })