polygram 0.12.0-rc.1 → 0.12.0-rc.2

package/lib/process-manager.js

@@ -123,6 +123,19 @@ const CALLBACK_TO_EVENT = {
   // menu auto-dismissed by `_waitForReady`. Surfacing the event so
   // soak can count how often aged-session resumes hit this path.
   onSessionAgePromptDismissed:  'session-age-prompt-dismissed',
+  // 0.12 CliProcess observability — typed hook events from cli-process.js
+  // _handleHookEvent. Each gets its own callback so polygram can persist
+  // structured rows to the events DB for soak-time aggregate queries.
+  //   - hook-lag-sample: Phase 1.8 — per-event lag_ms (target: median<2s, p99<5s)
+  //   - tool-result:     Phase 1.3 — PostToolUse durationMs per tool
+  //   - subagent-start / subagent-done: Phase 1.3 — typed subagent lifecycle
+  //     (we DO get tool-use='Agent' via onToolUse, but agent_type + durationMs
+  //      only fire on these typed events). SDK backend never emits — hooks
+  //     are CliProcess-specific (and were tmux-specific in 0.10–0.11).
+  onHookLagSample:              'hook-lag-sample',
+  onToolResult:                 'tool-result',
+  onSubagentStart:              'subagent-start',
+  onSubagentDone:               'subagent-done',
 };
 
 class ProcessManager {

package/lib/sdk/callbacks.js

@@ -622,6 +622,93 @@ function createSdkCallbacks({
       }
     },
 
+    // 0.12 Phase 1.8 — hook-lag persistence for the soak gate (median<2s,
+    // p99<5s). Each row carries the hookEventName + lagMs so we can:
+    //   SELECT json_extract(detail_json, '$.hook_event_name') AS evt,
+    //          AVG(json_extract(detail_json, '$.lag_ms')) AS avg_lag,
+    //          MAX(json_extract(detail_json, '$.lag_ms')) AS max_lag
+    //   FROM events WHERE kind='hook-lag-sample' AND ts>...
+    //   GROUP BY evt;
+    onHookLagSample: (sessionKey, payload /* , entry */) => {
+      try {
+        logEvent('hook-lag-sample', {
+          chat_id: getChatIdFromKey(sessionKey),
+          session_key: sessionKey,
+          backend: payload?.backend ?? 'cli',
+          hook_event_name: payload?.hookEventName ?? null,
+          lag_ms: payload?.lagMs ?? null,
+          tool_name: payload?.toolName ?? null,
+        });
+      } catch (err) {
+        logger.error?.(`[${botName}] hook-lag-sample handler: ${err.message}`);
+      }
+    },
+
+    // 0.12 Phase 1.3 — tool-result with durationMs. Pairs with the
+    // existing onToolUse row (which fires on PreToolUse) so the soak can
+    // compute per-tool average + p99 durations:
+    //   SELECT json_extract(detail_json, '$.tool_name') AS tool,
+    //          AVG(json_extract(detail_json, '$.duration_ms')) AS avg_ms,
+    //          MAX(json_extract(detail_json, '$.duration_ms')) AS max_ms
+    //   FROM events WHERE kind='tool-result' GROUP BY tool;
+    // isError captures the rare PostToolUse where the tool itself failed
+    // (vs the tool succeeding but claude deciding to retry).
+    onToolResult: (sessionKey, payload /* , entry */) => {
+      try {
+        logEvent('tool-result', {
+          chat_id: getChatIdFromKey(sessionKey),
+          session_key: sessionKey,
+          backend: payload?.backend ?? 'cli',
+          tool_name: payload?.name ?? null,
+          duration_ms: payload?.durationMs ?? null,
+          agent_id: payload?.agentId ?? null,
+          agent_type: payload?.agentType ?? null,
+          tool_use_id: payload?.toolUseId ?? null,
+          is_error: payload?.isError === true,
+        });
+      } catch (err) {
+        logger.error?.(`[${botName}] tool-result handler: ${err.message}`);
+      }
+    },
+
+    // 0.12 Phase 1.3 — subagent lifecycle. PreToolUse with name='Agent'
+    // synthesizes 'subagent-start' (no agent_id yet — claude doesn't
+    // hand one out until the inner SubagentStop). 'subagent-done' carries
+    // the agent_id + duration_ms so a soak can correlate the pair:
+    //   SELECT s.detail_json AS start, d.detail_json AS done
+    //   FROM events s JOIN events d
+    //     ON json_extract(s.detail_json, '$.tool_use_id') =
+    //        json_extract(d.detail_json, '$.tool_use_id')
+    //   WHERE s.kind='subagent-start' AND d.kind='subagent-done';
+    onSubagentStart: (sessionKey, payload /* , entry */) => {
+      try {
+        logEvent('subagent-start', {
+          chat_id: getChatIdFromKey(sessionKey),
+          session_key: sessionKey,
+          backend: payload?.backend ?? 'cli',
+          agent_type: payload?.agentType ?? null,
+          tool_use_id: payload?.toolUseId ?? null,
+        });
+      } catch (err) {
+        logger.error?.(`[${botName}] subagent-start handler: ${err.message}`);
+      }
+    },
+
+    onSubagentDone: (sessionKey, payload /* , entry */) => {
+      try {
+        logEvent('subagent-done', {
+          chat_id: getChatIdFromKey(sessionKey),
+          session_key: sessionKey,
+          backend: payload?.backend ?? 'cli',
+          agent_type: payload?.agentType ?? null,
+          agent_id: payload?.agentId ?? null,
+          duration_ms: payload?.durationMs ?? null,
+        });
+      } catch (err) {
+        logger.error?.(`[${botName}] subagent-done handler: ${err.message}`);
+      }
+    },
+
     onInjectFail: (sessionKey, payload /* , entry */) => {
       try {
         const msgId = payload?.msgId;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "polygram",
-  "version": "0.12.0-rc.1",
+  "version": "0.12.0-rc.2",
   "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": {