polygram 0.33.4 → 0.34.0

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://anthropic.com/claude-code/plugin.schema.json",
   "name": "polygram",
-  "version": "0.33.4",
+  "version": "0.34.0",
   "description": "Telegram integration for Claude Code that preserves the OpenClaw per-chat session model. Migration target for OpenClaw users. Multi-bot, multi-chat, per-topic isolation; SQLite transcripts; inline-keyboard approvals. Bundles /polygram:status|logs|pair-code|approvals admin commands plus history (transcript queries) and polygram-send (out-of-turn IPC sends with file-upload validation) skills.",
   "keywords": [
     "telegram",

package/lib/db/codex-reconciliation.js

@@ -152,6 +152,10 @@ function listUnresolvedCodexAttempts(rawDb, input = {}) {
          location.telegram_message_id,
          unresolved.telegram_source_message_id
        )
+     WHERE COALESCE(
+       location.telegram_message_id,
+       unresolved.telegram_source_message_id
+     ) IS NOT NULL
      ORDER BY unresolved.ambiguous_ts,
               unresolved.attempt_id
   `).all(sessionKey, sessionKey);

package/lib/handlers/dispatcher.js

@@ -59,7 +59,7 @@ function createDispatcher({
   // the historic 4096 for back-compat in synthetic test runs that pass
   // pre-formatted text.
   chunkBudget = 4096,
-  // Delay before a silent startup auto-retry re-dispatches (TMUX_SESSION_GONE).
+  // Delay before a silent startup auto-retry re-dispatches a pre-delivery failure.
   // Injected so tests can drive it to 0; production uses STARTUP_RETRY_DELAY_MS.
   startupRetryDelayMs = STARTUP_RETRY_DELAY_MS,
   // AUTH_DISABLED dedupe/re-arm gate (docs/AUTH_DISABLED_HANDLING_SPEC.md).
@@ -306,6 +306,11 @@ function createDispatcher({
         chat_id: chatId, session_key: sessionKey,
         msg_id: msg?.message_id,
         error: err.message?.slice(0, 500),
+        code: err.code || undefined,
+        cause_code: err.cause?.code || undefined,
+        stderr_tail: err.code === 'TMUX_SPAWN_FAILED' && typeof err.stderr === 'string'
+          ? err.stderr.replace(/[\x00-\x08\x0b-\x1f\x7f]/g, '').slice(-500)
+          : undefined,
         stack: err.stack?.split('\n').slice(0, 5).join('\n'),
         aborted: wasAborted || undefined,
         replay: isReplay || undefined,
@@ -442,23 +447,17 @@ function createDispatcher({
           }
         }
         const provider = providerRecovery?.provider;
-        // Startup auto-retry (option a, 2026-06-04). TMUX_SESSION_GONE = claude
-        // exited INSIDE the startup gate, before the dev-channels channel went
-        // live — so the user's message was NEVER delivered to claude. That makes
-        // a re-send idempotent BY CONSTRUCTION (unlike a mid-turn drop, where
-        // claude might still be slowly processing). The session_id was just
-        // poison-cleared above, so re-dispatching the SAME message spawns a FRESH
-        // session and delivers it. Silent: a transient startup flake (recurs
-        // ~once/9h on the channels backend) never reaches the user — instead of
-        // the "🔄 reset it, resend" papercut, polygram just retries. One-shot
-        // (_startupRetried) so a host that genuinely can't start claude surfaces
-        // the friendly reset reply (below) after EXACTLY one retry, never a loop.
-        // Scoped to TMUX_SESSION_GONE only: CHANNELS_DIALOG_TIMEOUT is a real
-        // blocking dialog (usage-limit / permission) a retry would just re-hit,
-        // so it keeps its "please resend" copy.
-        if (err.code === 'TMUX_SESSION_GONE' && !msg._startupRetried) {
+        // Retry once when startup fails before the bridge can deliver the user's
+        // message. TMUX_SESSION_GONE is poison-cleared above; TMUX_SPAWN_FAILED
+        // is already torn down by CliProcess's start-failure cleanup. Mid-turn
+        // failures and blocking dialogs are excluded because replay could
+        // duplicate work or immediately re-hit the same user decision.
+        const startupRetryable = err.code === 'TMUX_SESSION_GONE'
+          || err.code === 'TMUX_SPAWN_FAILED';
+        if (startupRetryable && !msg._startupRetried) {
           logEvent('startup-auto-retry', {
             chat_id: chatId, session_key: sessionKey, msg_id: msg?.message_id,
+            code: err.code,
           });
           // Re-dispatch a COPY carrying the one-shot marker — never mutate the
           // caller's msg (the boot-replay path shares/re-reads it). unref the

package/lib/ops/systemd-invocation.js

@@ -0,0 +1,22 @@
+'use strict';
+
+function parseSystemdInvocationId(env = process.env) {
+  const value = env.INVOCATION_ID;
+  if (value == null || value === '') return null;
+  if (!/^[0-9a-f]{32}$/.test(value)) {
+    throw new Error('invalid systemd invocation identity');
+  }
+  return value;
+}
+
+function lifecycleDetail(detail, invocationId) {
+  const value = detail && typeof detail === 'object' ? detail : {};
+  return invocationId == null
+    ? value
+    : { ...value, invocation_id: invocationId };
+}
+
+module.exports = {
+  parseSystemdInvocationId,
+  lifecycleDetail,
+};

package/ops/README.md

@@ -109,6 +109,19 @@ canonical absolute directory. Temporary paths and unsafe aliases are rejected;
 there is no `/tmp` fallback. If the data-directory path is too long for a
 portable Unix socket, set `POLYGRAM_IPC_DIR` to a shorter owner-only path.
 
+Deployment automation can query the aggregate in-flight handler count without
+printing session identifiers or the socket path:
+
+```sh
+POLYGRAM_IPC_DIR=/absolute/data-dir/.ipc polygram-ipc admin-bot busy
+# {"bot":"admin-bot","in_flight":0}
+```
+
+This is an advisory quiet-window check. The daemon's shutdown admission,
+delivery, and handler-settlement barriers remain the correctness boundary.
+Invoking `polygram-ipc admin-bot` without a subcommand retains the existing
+path-and-ping health output.
+
 Allowed methods: `sendMessage`, `sendPhoto`, `sendDocument`, `sendSticker`,
 `sendChatAction`, `editMessageText`, `setMessageReaction`. Other methods
 are rejected server-side.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polygram",
-  "version": "0.33.4",
+  "version": "0.34.0",
   "description": "Telegram daemon for Claude Code that preserves the OpenClaw per-chat session model. Migration path for OpenClaw users moving to Claude Code.",
   "main": "lib/ipc/client.js",
   "bin": {

package/polygram.js

@@ -32,6 +32,10 @@ const { buildPrompt, resolvePromptBackend } = require('./lib/prompt');
 const { countInFlight } = require('./lib/queue-utils');
 const { createIpcHandlers } = require('./lib/ipc/handlers');
 const { classifyOrphanSweep } = require('./lib/ops/tmux-preflight');
+const {
+  parseSystemdInvocationId,
+  lifecycleDetail,
+} = require('./lib/ops/systemd-invocation');
 const { filterAttachments, resolveFileCaps, resolveMaxFileOverride, MAX_TOTAL_BYTES } = require('./lib/attachments');
 // 0.9.0: SDK ProcessManager is the only pm. CLI pm
 // (lib/process-manager.js) deleted in commit 6.
@@ -3391,6 +3395,7 @@ let startPollWatchdog = null;
 
 async function main() {
   const cgroupOomObserver = createCgroupOomObserver();
+  const invocationId = parseSystemdInvocationId();
   let codexManagerOptions = Object.freeze({
     codexHostIdentity: null,
     codexBootSessionIdentity: null,
@@ -3628,7 +3633,10 @@ async function main() {
       console.log(`[inbox] swept ${swept.swept} files (${(swept.bytes / 1_048_576).toFixed(1)} MiB) older than ${inboxRetentionMs / 86_400_000}d`);
       db.logEvent('inbox-swept', { files: swept.swept, bytes: swept.bytes, retention_days: inboxRetentionMs / 86_400_000 });
     }
-    db.logEvent('polygram-start', { migration: migration.reason, imported: migration.imported });
+    db.logEvent('polygram-start', lifecycleDetail({
+      migration: migration.reason,
+      imported: migration.imported,
+    }, invocationId));
     if (cgroupOomObserver.startup.status === 'unavailable') {
       console.warn(`[oom-observer] unavailable (${cgroupOomObserver.startup.reason}); handled shutdown signals will retain clean-restart behavior`);
       db.logEvent('oom-observer-unavailable', { reason: cgroupOomObserver.startup.reason });
@@ -4534,7 +4542,7 @@ async function main() {
             });
           }
         }
-        logEvent('shutdown-drain', {
+        logEvent('shutdown-drain', lifecycleDetail({
           bot: BOT_NAME,
           in_flight: remaining,
           in_flight_at_signal: inFlightAtSignal,
@@ -4543,7 +4551,7 @@ async function main() {
           clean: res.clean,
           shutdown_reason: res.shutdownReason,
           ...(res.oomKillDelta != null ? { oom_kill_delta: res.oomKillDelta } : {}),
-        });
+        }, invocationId));
         console.log(`[shutdown] ${res.clean ? 'clean' : 'crash-like'} shutdown recorded (${res.shutdownReason}); ${inFlightAtSignal} in-flight at signal, drained ${drainElapsed}ms, ${remaining} still in-flight, ${res.replayMarked} marked replay-pending`);
       } catch (err) {
         console.error(`[shutdown] persistence failed: ${err.message}`);
@@ -4576,7 +4584,10 @@ async function main() {
     if (cancelAllWaiters) cancelAllWaiters('cancelled', 'polygram shutting down');
     if (pm && !pmRetired) await pm.shutdown().catch(() => {});
     if (db) {
-      try { db.logEvent('polygram-stop'); db.raw.close(); } catch {}
+      try {
+        db.logEvent('polygram-stop', lifecycleDetail({}, invocationId));
+        db.raw.close();
+      } catch {}
     }
     // rc.50: release our PID file claim so the next boot doesn't try
     // to kill us. releasePidFile is idempotent and only deletes the

package/scripts/ipc-smoke.js

@@ -1,19 +1,44 @@
 #!/usr/bin/env node
 /**
  * Quick IPC round-trip probe.
- * Usage: node scripts/ipc-smoke.js <bot-name>
+ * Usage: node scripts/ipc-smoke.js <bot-name> [busy]
  */
 
-const { call, socketPathFor } = require('../lib/ipc/client');
+const { call, readSecret, socketPathFor } = require('../lib/ipc/client');
 
 (async () => {
   const bot = process.argv[2] || 'shumabit';
+  const command = process.argv[3] || 'ping';
   const path = socketPathFor(bot);
 
+  if (command === 'busy') {
+    const result = await call({
+      path,
+      op: 'busy',
+      secret: readSecret(bot),
+    });
+    if (
+      result.ok !== true
+      || result.bot !== bot
+      || !Number.isSafeInteger(result.in_flight)
+      || result.in_flight < 0
+    ) {
+      throw new Error('invalid busy response');
+    }
+    console.log(JSON.stringify({
+      bot: result.bot,
+      in_flight: result.in_flight,
+    }));
+    return;
+  }
+
   console.log('path:', path);
   console.log('ping:', JSON.stringify(await call({ path, op: 'ping' })));
   console.log('DONE');
 })().catch((err) => {
-  console.error('ERR:', err.message);
+  const message = process.argv[3] === 'busy'
+    ? 'busy request failed'
+    : err.message;
+  console.error('ERR:', message);
   process.exit(1);
 });