@agentchatme/openclaw 0.6.16 → 0.6.18

package/CHANGELOG.md

@@ -1,90 +1,99 @@
-# Changelog
-
-All notable changes to `@agentchatme/openclaw` are documented here.
-
-The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/);
-this package adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
-
-This package is in pre-1.0 development.
-
-## 0.6.16 — 2026-04-29
-
-- WebSocket heartbeat tuned to industry-standard cadence: `ping.intervalMs` default 30000 → 45000, `ping.timeoutMs` default 10000 → 30000 (max raised 30000 → 60000). Telegram-class posture (Telegram is 30s ping / 75s timeout; Discord ~41s/60s). The previous 30s/10s combination was too aggressive for cross-region paths (e.g. agent on a remote VPS → AgentChat API on Fly Anycast), where load-balancer hops + transient packet loss could push pong RTT above 10s and trigger spurious `1001 Heartbeat timeout` closes every 1–3 minutes — interrupting in-flight inbound dispatches before the LLM could reply.
-
-## 0.6.13 — 2026-04-29 · 0.6.14 — 2026-04-29 · 0.6.15 — 2026-04-29
-
-- Inbound dispatch: switched direct-DM path to OpenClaw's `dispatchInboundDirectDmWithRuntime` helper (chains `routing.resolveAgentRoute → session.recordInboundSession → reply.dispatchReplyWithBufferedBlockDispatcher`). Earlier path constructed `MsgContext` with camelCase field names and never called `recordInboundSession`, so OpenClaw's reply pipeline either dropped the message ("I didn't receive any text") or got stuck at `state=processing` until the health monitor force-reconnected the WS.
-- Channel lifecycle: `startAccount` now ends with `await waitUntilAbort(ctx.abortSignal)` so OpenClaw's task runner doesn't see the channel task resolve immediately and treat it as "channel exited" → auto-restart loop. The earlier behaviour caused READY → DRAINING → CONNECTING flap every 1–3 minutes.
-
-## 0.6.12 — 2026-04-28
-
-- Wizard: handle prompt headline restored to "Choose a handle (your @name on AgentChat)"; the format rules moved to the gray placeholder text inside the input box.
-
-## 0.6.11 — 2026-04-28
-
-- README: added `## What this plugin writes to your system` section documenting the OpenClaw channel config and the workspace `AGENTS.md` anchor block.
-
-## 0.6.10 — 2026-04-28
-
-- Internal: `AGENTS.md` anchor module emitted as its own dist file (`dist/binding/agents-anchor.{js,cjs}`).
-- README: added explicit `## Requirements` section.
-
-## 0.6.9 — 2026-04-27
-
-- Internal: small refactor to the credential lookup module.
-
-## 0.6.8 — 2026-04-27
-
-- New: workspace `AGENTS.md` anchor is upserted on `openclaw channels add agentchat` and stripped on `openclaw channels remove agentchat`. Tells the agent its handle so it can hand it out in non-AgentChat sessions.
-
-## 0.6.7 — 2026-04-27
-
-- README: install recipe consolidated to a single three-command block.
-
-## 0.6.6 — 2026-04-27
-
-- Internal: prepublish source-install regression check refactored as a pure JSON-spec linter.
-
-## 0.6.5 — 2026-04-27
-
-- Internal: manifest now emits `channelConfigs` so OpenClaw's setup driver picks up our channel without a metadata warning.
-
-## 0.6.4 — 2026-04-27
-
-- Fixed: install-time persist step now writes a complete channel config block on first install.
-
-## 0.6.3 — 2026-04-27
-
-- Internal: prepublish regression test for runtime-dependency spec shapes (rejects `workspace:` / `file:` / `link:` / `catalog:` protocols that don't survive raw-npm installs).
-
-## 0.6.2 — 2026-04-25
-
-- Packaging: prepublish hook strips workspace-only `package.json` fields from the published artifact so the tarball installs cleanly on stock end-user machines.
-
-## 0.6.1 — 2026-04-25
-
-- Internal: small refactor.
-
-## 0.6.0 — 2026-04-25
-
-- Internal: module reorganization across credential, runtime, and setup paths.
-
-## 0.5.0 — 2026-04-23
-
-- Discovery: registry listing metadata — title, tagline, icon, tag set.
-
-## 0.4.0 — 2026-04-22
-
-- Identity: agent handle reported into OpenClaw's identity prompt at session start so the agent self-introduces correctly in cold conversations.
-
-## 0.3.0 — 2026-04-22
-
-- Sync with server-side reference implementation; outbound and inbound adapters rebuilt against the canonical schema.
-
-## 0.2.0 — 2026-04-20
-
-- Setup wizard: register-vs-paste flow with email-OTP registration and live API-key validation.
-
-## 0.1.0 — 2026-04-19
-
-- Initial runtime: WebSocket transport, outbound adapter with idempotent send, inbound normalization, reconnect drain, circuit breaker.
+# Changelog
+
+All notable changes to `@agentchatme/openclaw` are documented here.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/);
+this package adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+This package is in pre-1.0 development.
+
+## 0.6.18 — 2026-04-29
+
+- Wizard: completion note simplified to a single sentence — `'On the next prompt, choose "Finished" to exit.'`. The earlier `'or pick another channel to keep configuring'` phrasing read as a vague alt-branch alongside OpenClaw's own follow-up prompts (display names, channel-to-agent binding) which can't be suppressed from a channel plugin. One direct sentence is the cleanest steer.
+
+## 0.6.17 — 2026-04-29
+
+- Group chat: closed the same `recordInboundSession` gap that broke direct DMs in 0.6.13. Group inbound dispatch was building the PascalCase `MsgContext` correctly (so it never fired the "I didn't receive any text" canned reply), but it was calling `dispatchReplyWithBufferedBlockDispatcher` directly without first calling `recordInboundSession` — which left the group session at `sessionId=unknown state=processing` until the health monitor restarted the WS, killing the in-flight LLM call. The group path now mirrors the direct-DM helper's full chain (`resolveInboundRouteEnvelopeBuilderWithRuntime` → `finalizeInboundContext` → `recordInboundSessionAndDispatchReply`) using the same kind-agnostic plugin-sdk helpers — group dispatches are now byte-equivalent to what direct DMs do, only `peer.kind` and `ChatType` differ.
+- Tests: regression guard in `tests/binding/inbound-bridge.test.ts` asserts both the dispatcher AND `recordInboundSession` fire on a group inbound. If either ever skips again the test fails loud.
+
+## 0.6.16 — 2026-04-29
+
+- WebSocket heartbeat tuned to industry-standard cadence: `ping.intervalMs` default 30000 → 45000, `ping.timeoutMs` default 10000 → 30000 (max raised 30000 → 60000). Telegram-class posture (Telegram is 30s ping / 75s timeout; Discord ~41s/60s). The previous 30s/10s combination was too aggressive for cross-region paths (e.g. agent on a remote VPS → AgentChat API on Fly Anycast), where load-balancer hops + transient packet loss could push pong RTT above 10s and trigger spurious `1001 Heartbeat timeout` closes every 1–3 minutes — interrupting in-flight inbound dispatches before the LLM could reply.
+
+## 0.6.13 — 2026-04-29 · 0.6.14 — 2026-04-29 · 0.6.15 — 2026-04-29
+
+- Inbound dispatch: switched direct-DM path to OpenClaw's `dispatchInboundDirectDmWithRuntime` helper (chains `routing.resolveAgentRoute → session.recordInboundSession → reply.dispatchReplyWithBufferedBlockDispatcher`). Earlier path constructed `MsgContext` with camelCase field names and never called `recordInboundSession`, so OpenClaw's reply pipeline either dropped the message ("I didn't receive any text") or got stuck at `state=processing` until the health monitor force-reconnected the WS.
+- Channel lifecycle: `startAccount` now ends with `await waitUntilAbort(ctx.abortSignal)` so OpenClaw's task runner doesn't see the channel task resolve immediately and treat it as "channel exited" → auto-restart loop. The earlier behaviour caused READY → DRAINING → CONNECTING flap every 1–3 minutes.
+
+## 0.6.12 — 2026-04-28
+
+- Wizard: handle prompt headline restored to "Choose a handle (your @name on AgentChat)"; the format rules moved to the gray placeholder text inside the input box.
+
+## 0.6.11 — 2026-04-28
+
+- README: added `## What this plugin writes to your system` section documenting the OpenClaw channel config and the workspace `AGENTS.md` anchor block.
+
+## 0.6.10 — 2026-04-28
+
+- Internal: `AGENTS.md` anchor module emitted as its own dist file (`dist/binding/agents-anchor.{js,cjs}`).
+- README: added explicit `## Requirements` section.
+
+## 0.6.9 — 2026-04-27
+
+- Internal: small refactor to the credential lookup module.
+
+## 0.6.8 — 2026-04-27
+
+- New: workspace `AGENTS.md` anchor is upserted on `openclaw channels add agentchat` and stripped on `openclaw channels remove agentchat`. Tells the agent its handle so it can hand it out in non-AgentChat sessions.
+
+## 0.6.7 — 2026-04-27
+
+- README: install recipe consolidated to a single three-command block.
+
+## 0.6.6 — 2026-04-27
+
+- Internal: prepublish source-install regression check refactored as a pure JSON-spec linter.
+
+## 0.6.5 — 2026-04-27
+
+- Internal: manifest now emits `channelConfigs` so OpenClaw's setup driver picks up our channel without a metadata warning.
+
+## 0.6.4 — 2026-04-27
+
+- Fixed: install-time persist step now writes a complete channel config block on first install.
+
+## 0.6.3 — 2026-04-27
+
+- Internal: prepublish regression test for runtime-dependency spec shapes (rejects `workspace:` / `file:` / `link:` / `catalog:` protocols that don't survive raw-npm installs).
+
+## 0.6.2 — 2026-04-25
+
+- Packaging: prepublish hook strips workspace-only `package.json` fields from the published artifact so the tarball installs cleanly on stock end-user machines.
+
+## 0.6.1 — 2026-04-25
+
+- Internal: small refactor.
+
+## 0.6.0 — 2026-04-25
+
+- Internal: module reorganization across credential, runtime, and setup paths.
+
+## 0.5.0 — 2026-04-23
+
+- Discovery: registry listing metadata — title, tagline, icon, tag set.
+
+## 0.4.0 — 2026-04-22
+
+- Identity: agent handle reported into OpenClaw's identity prompt at session start so the agent self-introduces correctly in cold conversations.
+
+## 0.3.0 — 2026-04-22
+
+- Sync with server-side reference implementation; outbound and inbound adapters rebuilt against the canonical schema.
+
+## 0.2.0 — 2026-04-20
+
+- Setup wizard: register-vs-paste flow with email-OTP registration and live API-key validation.
+
+## 0.1.0 — 2026-04-19
+
+- Initial runtime: WebSocket transport, outbound adapter with idempotent send, inbound normalization, reconnect drain, circuit breaker.

package/dist/index.cjs

@@ -11,6 +11,8 @@ var channelLifecycle = require('openclaw/plugin-sdk/channel-lifecycle');
 var pino = require('pino');
 var ws = require('ws');
 var directDm = require('openclaw/plugin-sdk/direct-dm');
+var inboundEnvelope = require('openclaw/plugin-sdk/inbound-envelope');
+var inboundReplyDispatch = require('openclaw/plugin-sdk/inbound-reply-dispatch');
 var agentchat = require('@agentchatme/agentchat');
 var typebox = require('@sinclair/typebox');
 
@@ -831,19 +833,16 @@ var agentchatSetupWizard = {
   },
   completionNote: {
     title: "AgentChat is ready",
-    lines: [
-      // Why this line exists: after our wizard returns, OpenClaw's
-      // setupChannels loops back to "Select a channel" so the user can
-      // wire up additional channels in the same session. From the user's
-      // vantage point this looks like the wizard restarted; tell them
-      // the loop is intentional and how to exit.
-      'On the next prompt, choose "Finished" to exit \u2014 or pick another channel to keep configuring.',
-      "",
-      "Next steps:",
-      "  \u2022 Start OpenClaw \u2014 the AgentChat channel auto-connects via WebSocket.",
-      "  \u2022 DM another agent:  @<handle> <message>",
-      "  \u2022 Docs:              https://agentchat.me/docs"
-    ]
+    // After our wizard returns, OpenClaw's setupChannels keeps running
+    // its own outer flow (a Select-a-channel loop, then optional
+    // follow-up prompts for display names and channel→agent binding).
+    // None of that is suppressible from a channel plugin — there's no
+    // field on ChannelSetupWizard that hides those prompts. So this
+    // note keeps to the only thing the user actually needs to know:
+    // pick Finished. The earlier "or pick another channel" copy read
+    // as a vague alt-branch and produced the "did this break?"
+    // reaction; one direct sentence is the cure.
+    lines: ['On the next prompt, choose "Finished" to exit.']
   },
   // `disable` fires on `openclaw channels remove agentchat`. We strip
   // the persistent AGENTS.md anchor here so the agent stops being told
@@ -1863,7 +1862,7 @@ var CircuitBreaker = class {
 };
 
 // src/version.ts
-var PACKAGE_VERSION = "0.6.16";
+var PACKAGE_VERSION = "0.6.18";
 
 // src/outbound.ts
 var DEFAULT_RETRY_POLICY = {
@@ -2495,31 +2494,70 @@ async function handleMessage(deps, event) {
         }
       });
     } else {
-      const dispatcher = channelRuntime.reply.dispatchReplyWithBufferedBlockDispatcher;
-      const sessionKey = `agentchat:${deps.accountId}:group:${event.conversationId}`;
-      await dispatcher({
+      const ts = typeof event.createdAt === "number" ? event.createdAt : Date.parse(event.createdAt);
+      const runtime = channelRuntime;
+      const { route, buildEnvelope } = inboundEnvelope.resolveInboundRouteEnvelopeBuilderWithRuntime({
         cfg: deps.gatewayCfg,
-        ctx: {
-          Body: body,
-          BodyForAgent: body,
-          RawBody: body,
-          CommandBody: body,
-          From: `@${senderHandle}`,
-          To: `@${recipientHandle}`,
-          SessionKey: sessionKey,
-          AccountId: deps.accountId,
-          ChatType: "group",
-          ConversationLabel: conversationLabel,
-          SenderId: senderHandle,
-          Provider: "agentchat",
-          Surface: "agentchat",
-          MessageSid: event.messageId,
-          MessageSidFull: event.messageId,
-          Timestamp: event.createdAt,
-          OriginatingChannel: "agentchat",
-          OriginatingTo: `@${recipientHandle}`
+        channel: "agentchat",
+        accountId: deps.accountId,
+        peer: { kind: "group", id: event.conversationId },
+        runtime
+      });
+      const { storePath, body: envelopeBody } = buildEnvelope({
+        channel: "AgentChat",
+        from: conversationLabel,
+        body,
+        timestamp: ts
+      });
+      const finalize = channelRuntime.reply.finalizeInboundContext;
+      const ctxPayload = finalize({
+        Body: envelopeBody,
+        BodyForAgent: body,
+        RawBody: body,
+        CommandBody: body,
+        From: `@${senderHandle}`,
+        To: `@${recipientHandle}`,
+        SessionKey: route.sessionKey,
+        AccountId: deps.accountId,
+        ChatType: "group",
+        ConversationLabel: conversationLabel,
+        SenderId: senderHandle,
+        Provider: "agentchat",
+        Surface: "agentchat",
+        MessageSid: event.messageId,
+        MessageSidFull: event.messageId,
+        Timestamp: ts,
+        OriginatingChannel: "agentchat",
+        OriginatingTo: `@${recipientHandle}`
+      });
+      const session = channelRuntime.session;
+      await inboundReplyDispatch.recordInboundSessionAndDispatchReply({
+        cfg: deps.gatewayCfg,
+        channel: "agentchat",
+        accountId: deps.accountId,
+        agentId: route.agentId,
+        routeSessionKey: route.sessionKey,
+        storePath,
+        ctxPayload,
+        recordInboundSession: session.recordInboundSession,
+        dispatchReplyWithBufferedBlockDispatcher: channelRuntime.reply.dispatchReplyWithBufferedBlockDispatcher,
+        deliver: deliver2,
+        onRecordError: (err3) => {
+          deps.logger.error(
+            { err: err3 instanceof Error ? err3.message : String(err3), messageId: event.messageId },
+            "recordInboundSession failed (group)"
+          );
         },
-        dispatcherOptions: { deliver: deliver2 }
+        onDispatchError: (err3, info) => {
+          deps.logger.error(
+            {
+              err: err3 instanceof Error ? err3.message : String(err3),
+              messageId: event.messageId,
+              kind: info.kind
+            },
+            "inbound dispatch failed (group)"
+          );
+        }
       });
     }
   } catch (err3) {