@agfpd/iapeer 0.2.13 → 0.2.14

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agfpd/iapeer",
-  "version": "0.2.13",
+  "version": "0.2.14",
   "description": "Foundation core for the iapeer multi-agent ecosystem: identity, registry, storage, codec.",
   "type": "module",
   "bin": {

package/src/cli/cli.test.ts

@@ -7,7 +7,7 @@ import { afterEach, beforeEach, describe, expect, test } from 'bun:test'
 import { mkdtempSync, rmSync, writeFileSync } from 'fs'
 import { tmpdir } from 'os'
 import { join } from 'path'
-import { formatListTable, listPeers, parseArgs, removePeerCli, sendMessage, startPeer, stopPeer } from './index.ts'
+import { formatListTable, listPeers, parseArgs, removePeerCli, runCli, sendMessage, startPeer, stopPeer } from './index.ts'
 import { findPeer, readPeersIndex, upsertPeer } from '../registry/index.ts'
 import { isStopped, loadLifecycleConfig, setStopped } from '../lifecycle/index.ts'
 import { launchdPlistPath } from '../launch/launchd.ts'
@@ -132,6 +132,69 @@ describe('send validation', () => {
   })
 })
 
+describe('--help/-h global intercept (CLI hygiene — usage printed, NOTHING executed)', () => {
+  let captured: string
+  let origWrite: typeof process.stdout.write
+  beforeEach(() => {
+    captured = ''
+    origWrite = process.stdout.write
+    process.stdout.write = ((s: string | Uint8Array) => {
+      captured += typeof s === 'string' ? s : Buffer.from(s).toString('utf8')
+      return true
+    }) as typeof process.stdout.write
+  })
+  afterEach(() => {
+    process.stdout.write = origWrite
+  })
+
+  test('every verb with --help prints usage to stdout and exits 0', async () => {
+    // The full verb surface — including verbs with REAL side effects (onboard ran on
+    // prod swallowing --help; stop would set a durable flag; remove would delete).
+    const verbs = [
+      'onboard', 'install', 'update', 'rollback', 'version', 'daemon', 'status',
+      'install-runtime', 'init', 'create', 'list', 'stop', 'start', 'remove', 'send',
+      'enable', 'attach', 'interrupt', 'compact', 'self-fresh', 'native-memory', 'run-infra',
+    ]
+    for (const v of verbs) {
+      captured = ''
+      const code = await runCli([v, '--help'], env())
+      expect({ verb: v, code }).toEqual({ verb: v, code: 0 })
+      expect(captured).toContain('usage: iapeer')
+    }
+  })
+  test('-h works like --help, anywhere on the line', async () => {
+    expect(await runCli(['stop', 'somebody', '-h'], env())).toBe(0)
+    expect(captured).toContain('usage: iapeer')
+  })
+  test('bare `iapeer --help` / `-h` / `help` print usage', async () => {
+    for (const a of [['--help'], ['-h'], ['help']]) {
+      captured = ''
+      expect(await runCli(a, env())).toBe(0)
+      expect(captured).toContain('usage: iapeer')
+    }
+  })
+  test('--help does NOT execute the verb: `stop <peer> --help` leaves no durable stop flag', async () => {
+    await register('helpcheck')
+    const e = env()
+    expect(await runCli(['stop', 'helpcheck', '--help'], e)).toBe(0)
+    expect(isStopped(loadLifecycleConfig(e), 'claude-helpcheck')).toBe(false)
+  })
+  test('--help does NOT execute the verb: `remove <peer> --help` keeps the registry record', async () => {
+    await register('keepme')
+    const e = env()
+    expect(await runCli(['remove', 'keepme', '--help'], e)).toBe(0)
+    expect(findPeer(readPeersIndex({ env: e }), 'keepme')).not.toBeNull()
+  })
+  test('version --help shows usage, not the version number', async () => {
+    expect(await runCli(['version', '--help'], env())).toBe(0)
+    expect(captured).toContain('usage: iapeer')
+    expect(captured.trim().split('\n').length).toBeGreaterThan(3) // usage, not a bare semver line
+  })
+  test('a literal "--help" value stays expressible via --key=--help (not intercepted)', () => {
+    expect(parseArgs(['boris', '--message=--help']).flags.message).toBe('--help')
+  })
+})
+
 describe('parseArgs (audit #27 — value beginning with --)', () => {
   test('--key=value preserves a value that starts with --', () => {
     expect(parseArgs(['send', 'boris', '--message=--look', '--topic=re: x']).flags).toMatchObject({

package/src/cli/index.ts

@@ -60,6 +60,11 @@ export interface PeerListing {
   last_active_runtime?: Runtime
   intelligence: Intelligence
   description: string
+  /** The peer's working directory (registry fact). Machine-readable so host-local
+   *  tooling (e.g. the memory provider's init/verify rendering doctrine into
+   *  <cwd>/.iapeer/) keys on the REGISTRY instead of copying the layout default —
+   *  a layout change must not silently strand consumers (iapeer-memory ask, 10.06). */
+  cwd: string
   runtimes: RuntimeStatus[]
 }
 
@@ -104,6 +109,7 @@ export function listPeers(opts: CliEnvOptions = {}): PeerListing[] {
       last_active_runtime: lastActive,
       intelligence: peer.intelligence,
       description: peer.description,
+      cwd: peer.cwd,
       runtimes,
     }
   })
@@ -351,6 +357,7 @@ const USAGE = `usage: iapeer <verb> [args]
   update [version] [--force]                                     pull latest (or an exact version) of @agfpd/iapeer from npm + restart the daemon
   rollback                                                       revert to the previous binary (.prev) + restart the daemon
   version | --version | -v                                       print the installed binary's version
+  help | --help | -h                                             print this usage (works appended to any verb; executes nothing)
   daemon [--install-plist]                                       run the host-wide HTTP-MCP router (launchd-held)
   onboard [--dry-run] [--infra <csv>] [--no-memory] [--memory <pkg>]   register the agfpd marketplace (+ infra runtimes; + default memory provider, default YES)
   status                                                         host snapshot: version, daemon health, memory slot (<provider> | none)
@@ -373,6 +380,17 @@ const USAGE = `usage: iapeer <verb> [args]
 
 export async function runCli(argv: string[], env: NodeJS.ProcessEnv = process.env): Promise<number> {
   const [verb, ...rest] = argv
+  // CLI hygiene (design «Onboard костяка» §CLI-гигиена): an explicit help request —
+  // `--help`/`-h` ANYWHERE on the line, or the bare `help` verb — prints usage and
+  // executes NOTHING. Checked on the RAW argv BEFORE the switch: parseArgs would
+  // bury `--help` in flags no case reads (a cold-start `onboard --help` EXECUTED on
+  // prod — idempotency saved it), and `-h` would land in positionals. Token-exact
+  // match is safe: the look-ahead parser never consumes a `--`-token as a value, so
+  // a LITERAL "--help" value is only expressible as `--key=--help` (not intercepted).
+  if (verb === 'help' || argv.includes('--help') || argv.includes('-h')) {
+    process.stdout.write(USAGE)
+    return 0
+  }
   const { positionals, flags } = parseArgs(rest)
   const out = (s: string) => process.stdout.write(s)
   const errOut = (s: string) => process.stderr.write(s)

package/src/cli/listTui.test.ts

@@ -10,6 +10,7 @@ function row(over: Partial<PeerListing>): PeerListing {
     default_runtime: 'claude',
     intelligence: 'artificial',
     description: '',
+    cwd: '/tmp/p',
     runtimes: [{ runtime: 'claude', status: 'asleep' }],
     ...over,
   }

package/src/lifecycle/index.ts

@@ -572,6 +572,26 @@ function sessionAlive(sock: string, identity: string): boolean {
   return tmux(sock, 'has-session', '-t', identity).ok
 }
 
+/** Death-class tag for a gone session (live case: iapeer-memory 10.06 — the WHOLE
+ *  tmux server died by SIGKILL-class and exits.log stayed empty, because the
+ *  pane-died hook needs a living tmux event loop). Two distinguishable classes:
+ *  - `session-gone` — the server on the socket still ANSWERS but the session is not
+ *    there: a pane died inside a living server → the pane-died hook had its chance,
+ *    exits.log should carry the cause.
+ *  - `server-dead` — the server itself is gone: the socket file is missing, or it
+ *    exists but nothing serves it (stale socket — the SIGKILL/OOM class). pane-died
+ *    could never fire, so the lifecycle.log line is the only durable trace. */
+export function classifyGoneSession(sock: string): { death: 'server-dead' | 'session-gone'; reason: string } {
+  if (!existsSync(sock)) {
+    return { death: 'server-dead', reason: 'tmux server gone (socket file missing)' }
+  }
+  // Ask the SERVER, not the session: any server-level command answering (exit 0)
+  // proves the server is alive and merely lost this session.
+  return tmux(sock, 'list-sessions').ok
+    ? { death: 'session-gone', reason: 'session gone, tmux server alive (exit cause should be in exits.log)' }
+    : { death: 'server-dead', reason: 'tmux server dead — stale socket (SIGKILL/OOM class; exits.log has no entry)' }
+}
+
 // ─────────────────────────────────────────────────────────────────────────────
 // System-prompt composition for a woken peer (delegates the jq doctrine-merge to
 // launch/composeSystemPrompt). The tmux launch + boot/ready + activity-proxy all
@@ -1007,8 +1027,13 @@ export function superviseTick(cfg: LifecycleConfig, deps: SuperviseDeps = {}): S
       }
       // Crash / self-close: NO marker written, NO eager relaunch — the peer stays
       // asleep and wakes FRESH lazily on the next message (resolveWakeMode branch 3a).
-      out.push({ identity: s.identity, action: 'reaped-gone', reason: 'session no longer live' })
-      trace({ identity: s.identity, action: 'reaped-gone', reason: 'session no longer live', outcome: 'fresh-next-msg' })
+      // The death-class tag (classifyGoneSession) makes the two gone-classes
+      // distinguishable in lifecycle.log: `session-gone` (pane died, server alive →
+      // exits.log should have the cause) vs `server-dead` (whole tmux server died →
+      // exits.log structurally empty; this line is the only durable trace).
+      const gone = classifyGoneSession(sock)
+      out.push({ identity: s.identity, action: 'reaped-gone', reason: gone.reason })
+      trace({ identity: s.identity, action: 'reaped-gone', death: gone.death, reason: gone.reason, outcome: 'fresh-next-msg' })
       continue
     }
     // Idle accounting via the runtime adapter's activity proxy (claude transcript

package/src/lifecycle/lifecycle.test.ts

@@ -4,6 +4,7 @@ import { tmpdir } from 'os'
 import { join } from 'path'
 import {
   attachPeer,
+  classifyGoneSession,
   clearEphemeralArmed,
   clearNewEager,
   clearStopped,
@@ -229,6 +230,8 @@ describe('superviseTick H4 guard', () => {
     const logged = readFileSync(join(c.eventLogDir, 'lifecycle.log'), 'utf8')
     expect(logged).toContain(`ev=supervise identity=${id} action=reaped-gone`)
     expect(logged).toContain('outcome=fresh-next-msg')
+    // death-class tag: no socket file at all in this sandbox → the server is gone
+    expect(logged).toContain('death=server-dead')
   })
 
   test('empty state dir → no outcomes', () => {
@@ -621,6 +624,83 @@ describe('ephemeral-armed marker + config', () => {
   })
 })
 
+// ─────────────────────────────────────────────────────────────────────────────
+// classifyGoneSession — the death-class tag for reaped-gone (server-dead vs
+// session-gone). Live case: iapeer-memory 10.06 — the whole tmux server died
+// (SIGKILL class), exits.log stayed empty; lifecycle.log must carry the class.
+// ─────────────────────────────────────────────────────────────────────────────
+
+describe('classifyGoneSession (death-class tag)', () => {
+  const tmuxAvailable = spawnSync('tmux', ['-V'], { stdio: 'ignore' }).status === 0
+
+  test('missing socket file → server-dead', () => {
+    const r = classifyGoneSession(join(tmpdir(), 'iapeer-no-such-sock-ever.sock'))
+    expect(r.death).toBe('server-dead')
+    expect(r.reason).toContain('socket file missing')
+  })
+
+  test('stale socket (file exists, nothing serves it) → server-dead', () => {
+    // The SIGKILL/OOM class: the killed server never unlinks its socket file.
+    const dir = mkdtempSync(join(tmpdir(), 'iapeer-stale-sock-'))
+    const sock = join(dir, 'tmux-iap-claude-stale.sock')
+    try {
+      writeFileSync(sock, '') // a plain file — tmux cannot connect to it
+      const r = classifyGoneSession(sock)
+      expect(r.death).toBe('server-dead')
+      expect(r.reason).toContain('stale socket')
+    } finally {
+      rmSync(dir, { recursive: true, force: true })
+    }
+  })
+
+  test.if(tmuxAvailable)('server alive but the session is not there → session-gone', () => {
+    const dir = mkdtempSync(join(tmpdir(), 'iapeer-sg-sock-'))
+    const sock = join(dir, 'tmux-iap-claude-sg.sock')
+    try {
+      // a LIVING server on the socket holding some OTHER session
+      spawnSync('tmux', ['-S', sock, 'new-session', '-d', '-s', 'other-session', 'sleep', '60'])
+      const r = classifyGoneSession(sock)
+      expect(r.death).toBe('session-gone')
+      expect(r.reason).toContain('server alive')
+    } finally {
+      spawnSync('tmux', ['-S', sock, 'kill-server'], { stdio: 'ignore' })
+      rmSync(dir, { recursive: true, force: true })
+    }
+  })
+
+  test.if(tmuxAvailable)('superviseTick logs death=session-gone when the pane died inside a living server', () => {
+    const root = mkdtempSync(join(tmpdir(), 'iapeer-sg-root-'))
+    const laDir = mkdtempSync(join(tmpdir(), 'iapeer-sg-la-')) // empty → not launchd-managed
+    const env = {
+      ...process.env,
+      IAPEER_ROOT: root,
+      IAPEER_LAUNCHAGENTS_DIR: laDir,
+      IAPEER_SOCK_DIR: join(root, 'socks'),
+    }
+    const cfg = loadLifecycleConfig(env)
+    const identity = 'claude-sg'
+    const sock = join(root, 'socks', 'tmux-iap-claude-sg.sock')
+    try {
+      mkdirSync(join(root, 'socks'), { recursive: true })
+      mkdirSync(cfg.stateDir, { recursive: true })
+      // the server LIVES (another session holds it) but claude-sg's session is gone
+      spawnSync('tmux', ['-S', sock, 'new-session', '-d', '-s', 'placeholder', 'sleep', '60'])
+      writeFileSync(
+        join(cfg.stateDir, `${identity}.session`),
+        JSON.stringify({ identity, runtime: 'claude', personality: 'sg', cwd: '/tmp/none', wokeAt: 0 }),
+      )
+      const out = superviseTick(cfg, { env })
+      expect(out.find(x => x.identity === identity)?.action).toBe('reaped-gone')
+      const logged = readFileSync(join(cfg.eventLogDir, 'lifecycle.log'), 'utf8')
+      expect(logged).toContain(`identity=${identity} action=reaped-gone death=session-gone`)
+    } finally {
+      spawnSync('tmux', ['-S', sock, 'kill-server'], { stdio: 'ignore' })
+      rmSync(root, { recursive: true, force: true })
+      rmSync(laDir, { recursive: true, force: true })
+    }
+  })
+})
+
 describe('superviseTick quiet-reap (M2 die-after-reply, real tmux)', () => {
   const tmuxAvailable = spawnSync('tmux', ['-V'], { stdio: 'ignore' }).status === 0
 

package/src/update/index.ts

@@ -226,7 +226,18 @@ export function updateIapeer(deps: UpdateDeps = {}): UpdateResult {
 
   const runInstall = deps.runInstall ?? defaultRunInstall
   if (!runInstall(desired, env)) {
-    return { status: 'failed', from, latest: desired, reason: `\`npx ${IAPEER_PACKAGE}@${desired} install\` failed` }
+    // NB: the installer is the DETERMINISTIC pack+build path (no npx — see
+    // defaultRunInstall); the most common cause right after a publish is the npm
+    // CDN tarball lagging the version metadata (live-hit 10.06: `npm view` already
+    // showed the version, `npm pack` still failed; a retry ~1 min later succeeded).
+    return {
+      status: 'failed',
+      from,
+      latest: desired,
+      reason:
+        `deterministic install of ${IAPEER_PACKAGE}@${desired} failed (npm pack/deps/build) — ` +
+        `if just published, the registry tarball may still be propagating; retry in ~1 min`,
+    }
   }
 
   const restart = deps.restartDaemon ?? kickstartDaemon