@lansenger-pm/openclaw-lansenger-channel 3.15.0 → 3.15.2

package/CHANGELOG.md

@@ -4,6 +4,112 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/), and this project adheres to [Semantic Versioning](https://semver.org/).
 
+## [3.15.2] - 2026-06-18
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **Slash commands produce no reply**: `ctxPayload.CommandAuthorized` was hardcoded to `undefined`, causing the agent core to skip command execution and treat `/reset`/`/new` as plain agent text. Now set to the actual `resolveChannelMessageIngress` authorization result.
+
+## [3.15.1] - 2026-06-18
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+>
+> Supersedes the deprecated `3.15.0` (published by mistake on 2026-06-04). Includes all fixes from `3.14.5` through `3.14.8`.
+
+### Bug Fixes (v3.14.5 → v3.14.8)
+
+- **Group chat reply routing**: Outbound adapters created fresh client instances with empty `chatTypeMap`, causing group replies to route via private-message API.
+- **`lansenger_send_*` tools multi-account routing**: Tools used `getRunningClient()` which always returned the first account's client. Changed to factory pattern with `ctx.agentAccountId`.
+- **Command authorization ignores global `commands.ownerAllowFrom`**: Slash commands blocked when owner configured globally rather than per-account.
+- **Multi-account audit**: Outbound adapters, status snapshot, reply fallback delivery, and approval card store all now use per-account lookup via `getRunningEntryByAccount()`.
+
+### Chore
+
+- Removed dead imports (`getRunningClient`, `getLastInboundTime`) from channel.ts after multi-account refactor.
+
+## [3.14.8] - 2026-06-18
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **Command authorization ignores global `commands.ownerAllowFrom`**: Slash commands (`/reset`, etc.) were blocked when `allowFrom` was only configured at the global `commands.ownerAllowFrom` level rather than per-account. Fixed by extracting `lansenger:`-prefixed entries from the global config and passing them as `commandOwnerAllowFrom` to `resolveChannelMessageIngress`.
+- **Multi-account audit — outbound adapters**: 7 outbound adapter methods (`sendText`, `sendMedia`, `sendFormattedText`, etc.) used `getRunningClient()` which returns the first running account's client. In multi-bot setups, this could deliver messages using the wrong account's credentials. Changed to `getRunningEntryByAccount(account.accountId)` to match the correct running entry.
+- **Multi-account audit — status snapshot**: `buildAccountSnapshot` used `getRunningClient()` for connected state and `getLastInboundTime()` for last activity, both returning data from the first running account instead of the account being probed. Added `getLastInboundTimeByAccount()` and switched to per-account lookup.
+- **Multi-account audit — reply fallback delivery**: The `reply_payload_sending` fallback hook used `getRunningClient()` with no account resolution, potentially delivering replies via the wrong bot. Now prefers per-account resolution via `getRunningEntryByAccount()`.
+- **Multi-account audit — approval card store collision**: `pendingApprovalCards` was keyed by chatId only, risking cross-account overwrites when different Lansenger instances shared chatIds. Now keyed by `accountId:chatId`.
+
+## [3.14.7] - 2026-06-18
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **`lansenger_send_*` tools multi-account routing**: All `lansenger_send_*` tools used `getRunningClient()` which always returned the first entry from `runningAccounts` Map, causing cross-account message sending in multi-bot setups. Fixed by:
+  - Added `getRunningEntryByAccount(accountId)` in runtime.ts that matches by account key/ID, with single-account fallback.
+  - Changed tool registration to use OpenClaw's tool factory pattern (`api.registerTool((ctx) => {...})`), resolving the correct client via `ctx.agentAccountId` from the OpenAI-compatible tool context.
+
+## [3.14.6] - 2026-06-17
+
+> **Hotfix**: v3.14.5 was published with a broken `dist/src/runtime.js` that imports `./persistent-store.js` but the file was missing.
+
+### Bug Fixes
+
+- **Missing `persistent-store.js` in npm bundle**: `src/persistent-store.ts` was uncommitted but referenced by modified `src/runtime.ts`. `prepublishOnly` ran `tsc` against the working directory, producing `dist/src/runtime.js` with the import, but the compiled `dist/src/persistent-store.js` was excluded from the npm tarball. Now committed and included.
+
+## [3.14.5] - 2026-06-17
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **Group chat reply routing**: Outbound adapters (`sendText`, `sendMedia`, `sendFormattedText`, `sendPayload`, `beforeDeliverPayload`) created fresh `LansengerClient` instances via `makeClient()`, which had an empty `chatTypeMap` cache. `isGroupChat()` then fell back to `chatId.startsWith("group:")` — which never matches Lansenger group IDs — causing all group replies to be sent via private-message API instead of group-message API. Fixed by using `getRunningClient() ?? makeClient()` so the cached client (with populated `chatTypeMap`) handles routing.
+
+## [3.14.4] - 2026-06-15
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **`recoverPendingInboundContexts` duplicate restart notice**: Gateway restart-recovery sent multiple "system restart" notices to the same chat due to two issues:
+  1. `startLansengerGateway` was called from both runtime extension and channel extension initialization paths, causing `recoverPendingInboundContexts` to execute twice. Added a module-level `recoveryGuard` flag for idempotency.
+  2. Even within a single invocation, multiple pending contexts for the same `chatId` would each trigger a separate notice. Added `chatId` deduplication — only one notice per chat per recovery cycle.
+- **`recoverPendingInboundContexts` duplicate ack revoke**: Same guard prevents the stale ack message from being revoked twice.
+- **WS reconnection without exponential backoff**: `backoffIdx` was reset to 0 immediately after `new WebSocket()` instead of on actual connection (`ws.onopen`). This caused every reconnect attempt to use a fixed 2s delay and log `attempt 1` regardless of retry count. Moved `this.backoffIdx = 0` into `ws.onopen` so the `[2, 5, 10, 30, 60]` backoff array is properly utilized.
+
+## [3.14.3] - 2026-06-04
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Features
+
+- **Gateway restart-recovery inbound context persistence**: Active inbound sessions are persisted to `~/.openclaw/lansenger-inbound-contexts.json` before `inbound.run`. On gateway restart/plugin upgrade, `recoverPendingInboundContexts` detects interrupted sessions, revokes stale ack messages, and sends a "system restart, reprocessing..." notification. Contexts older than 5 minutes are discarded.
+- **User notification**: Sends a locale-aware restart notice (`zh`/`en`) to the affected chat after recovery.
+
+## [3.14.2] - 2026-06-04
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Features
+
+- **`reply_payload_sending` fallback delivery**: Tracks active `inbound.run` sessions via `activeDeliverySessions` set. When `reply_payload_sending` fires for a session without an active `inbound.run` (e.g., after gateway restart/turn timeout), the reply is delivered directly via the Lansenger client instead of being silently dropped. Extracts `chatId` from `sessionKey` to resolve delivery target.
+
+## [3.14.1] - 2026-06-04
+
+> **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
+
+### Bug Fixes
+
+- **`ackMessage` default**: Changed from `false` to `true`; `revokeAckMessage` from `true` to `false`, aligning with the `after_agent_dispatch` ack policy.
+- **`pendingApprovalCards` persistence**: Migrated from in-memory `Map` to `~/.openclaw/lansenger-approval-cards.json` file-backed store, surviving gateway restarts.
+- **Group ingress fallback**: Added `requireMention` + `isAtMe` check to prevent non-@ messages from triggering the bot.
+- **`isPathAllowed` security hardening**: Empty `mediaLocalRoots` now defaults to `[cwd, tmpdir]` instead of allowing any path.
+- **Dead code removal**: Removed unused `debounceMs` statement in `runtime.ts`.
+- **`convertPxToPtDeep` → `convertPxToPtCard`**: Only transforms known style fields; skips URL/link fields to avoid corruption.
+- **Documentation**: All 5 locale READMEs updated with new defaults and `mediaLocalRoots` docs.
+
 ## [3.14.0] - 2026-06-04
 
 > **Compatible with OpenClaw `^2026.6.1`** (tested against `2026.6.1`).
@@ -25,60 +131,43 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/), and this
 - **Test coverage**: 4 new tests via `verifyChannelMessageReceiveAckPolicyAdapterProofs` and `listDeclaredReceiveAckPolicies`, verifying all declared policies pass SDK contract proofs.
 - This replaces the previously implicit ack timing embedded in the `runtime.ts` monitor-local state with a standardized, SDK-verifiable contract.
 
-## [3.13.0] - 2026-06-03
+## [3.13.1] - 2026-06-03
 
 > **Compatible with OpenClaw `^2026.5.28`** (tested against `2026.5.28`).
 
-### Bug Fix
+- migrate `message_sending` from `api.registerHook` to `api.on("message_sending", ...)` (OpenClaw 2026.5.28 rejects `registerHook` for typed hooks that lack a name)
+
+## [3.13.0] - 2026-06-03
 
-- **Missing `dist/src/setup-i18n.*` in npm package**: `setup-wizard.js` imports `./setup-i18n.js`, but the `files` field in `package.json` did not include `dist/src/setup-i18n.*`, causing `Cannot find module './setup-i18n.js'` at gateway startup. Added `dist/src/setup-i18n.*` to the `files` array.
-- **`message_sending` hook registration missing name**: `api.registerHook("message_sending", ...)` now fails at gateway startup with "hook registration missing name" (OpenClaw 2026.5.28 rejects `registerHook` calls for typed hooks that lack a name). Migrated `message_sending` to `api.on("message_sending", ...)` — the typed lifecycle hook API — alongside the existing `reply_payload_sending` hook.
+> **Compatible with OpenClaw `^2026.5.28`** (tested against `2026.5.28`).
 
-### Outbound Hook Expansion
+- Add `dist/src/setup-i18n.*` to `files` field in package.json (fixes `Cannot find module './setup-i18n.js'` at gateway startup)
 
-- **`reply_payload_sending` hook**: Register typed plugin hook via `api.on()` for the lansenger channel. Intercepts reply payloads before delivery — enables logging, approval context injection, and payload rewrite/cancel decisions.
-- **`message_sending` hook**: Register typed plugin hook via `api.on()` for early-stage reply interception.
-- **`normalizePayload` — code-block detection**: Payloads containing `\`\`\`` code fences are now marked `_lansengerFormatText: true` even when they also carry `mediaUrl` or `presentation`. Previously, code-rich agent replies with attachments fell back to plain-text `msgType`, losing Markdown formatting. Now they always render via `formatText`.
-- **`beforeDeliverPayload` — approval-resolved crash recovery**: When `hint.kind === "approval-resolved"`, the delivery pipeline now proactively updates the original appCard status via `updateDynamicCard`. Because the delivery pipeline replays on gateway restart, this provides crash-recovery safety.
-- **`shouldSuppressLocalPayloadPrompt`**: Suppresses the local text prompt when the native approval route is active (`hint.kind === "approval-pending"` with `nativeRouteActive === true`). Prevents duplicate approval prompts.
-- **`pendingApprovalCards` Map**: Tracks sent approval card `messageId`s keyed by `chatId`. Enables `beforeDeliverPayload` to correlate approval-resolved payloads with the original card for status updates.
-- **Text chunking**: Add `textChunkLimit: 4000`, `chunkerMode: "markdown"`, and `chunker` (using SDK `chunkMarkdownTextWithMode`). Long agent replies are automatically split into multiple Lansenger messages, preserving Markdown structure across chunks.
-- **`resolveEffectiveTextChunkLimit`**: Uses SDK `resolveTextChunkLimit` to allow config-level override of the chunk limit.
-- Add `normalizePayload` and `beforeDeliverPayload` callbacks on outbound base (structural preparation for outbound hooks).
-- Add session-scoped delivery dedup (`sessionDeliveryTracker`) to prevent duplicate sends across turns in the same session.
+## [3.12.2] - 2026-06-03
 
-### OpenClaw 2026.5.27–28 Compatibility
+> **Compatible with OpenClaw `^2026.5.28`** (tested against `2026.5.28`).
 
-- Migrate `api.runtime.channel.turn` → `api.runtime.channel.inbound` (OpenClaw removed the old `turn` runtime alias; the new `inbound` namespace provides the same `run` function with identical parameters). This is a **required** migration — the old `turn` alias no longer exists in OpenClaw 2026.5.27.
-- Thread canonical `sessionKey` into outbound hooks (`sendText`, `sendMedia`, `sendFormattedText`) for multi-session/multi-agent routing and dedup.
-- Fallback `sessionKey` now includes `accountId` for multi-bot scenarios (`agent:main:lansenger:<accountId>:<chatType>:<chatId>`).
-- Pin npm install spec to exact version (`@lansenger-pm/openclaw-lansenger-channel@3.13.0`) to prevent supply-chain attacks and accidental upgrades.
-- Bump `openclaw` devDependency from `^2026.5.20` to `^2026.5.28`.
-- Split changelog out of READMEs into standalone `CHANGELOG.md` (all 5 locale READMEs now reference this file).
+- **Outbound Hook Expansion**: `reply_payload_sending` hook, `message_sending` hook, `normalizePayload` code-block detection, `beforeDeliverPayload` approval-resolved crash recovery, `shouldSuppressLocalPayloadPrompt`, `pendingApprovalCards` Map, text chunking (`textChunkLimit: 4000`, `chunkerMode: "markdown"`), `resolveEffectiveTextChunkLimit`, session-scoped delivery dedup (`sessionDeliveryTracker`)
+- **Approval Flow Enhancement**: `approvalCapability.transport.send` stores card `messageId` in `pendingApprovalCards`; card status updates via both `transport.update` and `beforeDeliverPayload` (crash-recovery safety)
+- **Setup Wizard i18n**: `createSetupTranslator()` + plugin-owned dictionary, locales `en`/`zh-CN`/`zh-TW`, `src/setup-i18n.ts`
 
-### Approval Flow Enhancement
+## [3.12.1] - 2026-06-03
 
-- `approvalCapability.transport.send` now stores card `messageId` in `pendingApprovalCards` for later correlation in `beforeDeliverPayload`.
-- Card status updates (pending → approved/denied) now happen via both `transport.update` (approval framework) and `beforeDeliverPayload` (delivery pipeline). The latter provides crash-recovery safety.
+> **Compatible with OpenClaw `^2026.5.28`** (tested against `2026.5.28`).
 
-### Setup Wizard i18n
+- **Security**: SecretRef support for appSecret, plaintext appSecret warning, SSRF protection for `sendImageUrl`, `dangerouslyAllowPrivateNetwork` config, `mediaLocalRoots` config, path validation (`isPathAllowed`)
+- Pin npm install spec to exact version (`@lansenger-pm/openclaw-lansenger-channel@3.12.1`)
+- Thread canonical `sessionKey` into outbound hooks for multi-session/multi-agent routing
+- Fallback `sessionKey` includes `accountId` for multi-bot scenarios
+- Bump `openclaw` devDependency from `^2026.5.20` to `^2026.5.28`
 
-- **Locale-aware setup wizard**: Replace all bilingual hard-coded strings (`"Chinese / English"`) with `createSetupTranslator()` + plugin-owned dictionary. The wizard now displays in the user's locale (resolved from `OPENCLAW_LOCALE` → `LC_ALL` → `LC_MESSAGES` → `LANG`, fallback English).
-- Supported locales: **`en`**, **`zh-CN`**, **`zh-TW`**. French and other locales fall back to English for wizard prompts (READMEs remain 5-language).
-- Common strings (status labels, "Docs:" prefix, etc.) use the built-in OpenClaw catalog keys (`wizard.channels.*`). Plugin-specific strings use a local dictionary with `lt()`.
-- New file: `src/setup-i18n.ts` — locale resolver + 40+ entry dictionary.
+## [3.12.0] - 2026-06-03
 
-### Security
+> **Compatible with OpenClaw `^2026.5.27`** (tested against `2026.5.27`).
 
-- **SecretRef support for appSecret**: `resolveAccount()` now detects SecretRef objects via `coerceSecretRef()` and resolves from env vars. No need to store appSecret as plaintext in config. After running `openclaw secrets configure`, the config contains `__OPENCLAW_SECRET__({ref_id})` instead of the raw secret value, while the actual value is stored in the system credential store.
-- **Plaintext appSecret warning in `resolveAccount()`**: Emits `log.warn` on every gateway startup when `appSecret` is stored as a plaintext string in `openclaw.json`, advising migration to SecretRef via `openclaw secrets configure`.
-- **Plaintext appSecurity warning in setup wizard `finalize`**: Console output in the user's locale when plaintext appSecret is detected after config migration.
-- **`securityNote` in setup wizard**: A conditional note that appears only when a plaintext appSecret exists, with locale-appropriate migration instructions.
-- **README security advisory**: All 5 locale READMEs now include a prominent `⚠️ Security: Migrate appSecret to SecretRef storage` block.
-- **SSRF protection for sendImageUrl**: `assertHttpUrlTargetsPrivateNetwork()` blocks RFC1918/link-local/metadata-IP targets by default.
-- **`dangerouslyAllowPrivateNetwork` config**: Opt-in at top-level or per-account to allow private network image URLs. Audit finding (`lansenger/dangerously-allow-private-network`) warns when enabled.
-- **`mediaLocalRoots` config**: Restrict local file delivery to configured directories. Prevents agents from accessing arbitrary files outside allowed roots. Available at top-level and per-account.
-- **Path validation**: `isPathAllowed()` validates local file paths against `mediaLocalRoots` before delivery; blocked paths are logged and skipped.
+- **Breaking**: Migrate `api.runtime.channel.turn` → `api.runtime.channel.inbound` (OpenClaw removed the old `turn` runtime alias)
+- Split changelog out of READMEs into standalone `CHANGELOG.md`
+- All 5 locale READMEs now reference this file
 
 ## [3.11.0] - 2026-05-26
 
@@ -120,12 +209,15 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/), and this
 
 > **Compatible with OpenClaw `^2026.5.20`** (tested against `2026.5.20`).
 
-### Features
+- Fix apiGatewayUrl lint check to skip when accounts have gateway set.
 
-- Add `security.collectWarnings` and `security.collectAuditFindings` for `openclaw doctor --lint` integration. Checks: credentials missing/incomplete, dmPolicy not pairing, apiGatewayUrl not set, group config unused.
+## [3.8.1] - 2026-05-20
+
+> **Compatible with OpenClaw `^2026.5.20`** (tested against `2026.5.20`).
+
+- Add `security.collectWarnings` and `security.collectAuditFindings` for `openclaw doctor --lint` integration.
 - Add `doctor.repairConfig` to auto-fix dmPolicy to pairing.
 - Bump `openclaw` devDependency from `^2026.5.7` to `^2026.5.20`.
-- Fix apiGatewayUrl lint check to skip when accounts have gateway set.
 
 ## [3.7.9] - 2026-05-18
 
@@ -388,4 +480,4 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/), and this
 - Multi-language READMEs (en, zhHans, zhHant, zhHantHK, fr).
 - DM security: pairing mode (default), allowlist, open, disabled.
 - Approval workflow: appCard with `headStatusInfo` for pending/approved/denied states.
-- Auto-start WebSocket gateway on plugin activation.
+- Auto-start WebSocket gateway on plugin activation.

package/LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2026 Lansenger OpenClaw Channel
+Copyright (c) 2026 Lanxin Mobile (Beijing) Technology Co., ltd.
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

package/README.fr.md

@@ -12,7 +12,7 @@ Connecte OpenClaw à Lansenger — une plateforme de messagerie d'entreprise —
 ## Fonctionnalités
 
 - **Messagerie en temps réel** via connexion longue WebSocket
-- **Support multi-bot** — lier plusieurs bots Lansenger à différents agents OpenClaw
+- **Support multi-robot** — lier plusieurs robots Lansenger à différents agents OpenClaw
 - **Support Markdown** utilisant le msgType `formatText` (par défaut)
 - **Fichiers/Images/Vocaux** via le msgType `text` avec upload de médias
 - **Cartes d'approbation** — workflow d'approbation interactif avec mises à jour de statut en place (en attente → approuvé/refusé)
@@ -20,8 +20,8 @@ Connecte OpenClaw à Lansenger — une plateforme de messagerie d'entreprise —
 - **Auto-routage via msgTarget** — toutes les méthodes d'envoi routent automatiquement vers les API groupe ou DM (privé) ; pas de méthodes groupe/privé séparées
 - **@Mentions** — support @tout et @utilisateurs spécifiques dans les chats de groupe
 - **Traitement des médias entrants** — téléchargement d'images/fichiers/vocaux, détection d'extension, chemins de fichiers pour l'agent
-- **Révocation de messages** — révoquer les messages précédemment envoyés (chatType : bot ou groupe uniquement)
-- **Démarrage automatique** — la passerelle connecte automatiquement tous les comptes de bots configurés au démarrage
+- **Révocation de messages** — révoquer les messages précédemment envoyés (chatType : robot ou groupe uniquement)
+- **Démarrage automatique** — la passerelle connecte automatiquement tous les comptes de robots configurés au démarrage
 - **Debounce entrant** — fusionner les messages consécutifs rapides du même émetteur via la config OpenClaw `messages.inbound.debounceMs`
 - **Message ack** — envoyer un message « reçu, en cours de traitement... » avant le traitement de l'agent ; auto-révoqué après la réponse ; langue auto-détectée
 - **Zéro modification du core** — mode plugin pur, `git diff HEAD` reste INTACT
@@ -102,7 +102,7 @@ openclaw gateway restart
 
 ### Premier message
 
-Après le redémarrage, le bot se connecte automatiquement via WebSocket. Envoyez un DM au bot — vous recevrez un code de pairage. Approuvez-le :
+Après le redémarrage, le robot se connecte automatiquement via WebSocket. Envoyez un DM au robot — vous recevrez un code de pairage. Approuvez-le :
 
 ```bash
 openclaw pairing approve lansenger <code>
@@ -116,8 +116,8 @@ Ajoutez ces variables à `~/.openclaw/.env` ou à votre environnement :
 
 | Variable | Description | Exemple |
 |----------|-------------|---------|
-| `LANSENGER_APP_ID` | App ID du bot personnel | `your-appid` |
-| `LANSENGER_APP_SECRET` | App Secret du bot personnel | `ABCDEF123456...` |
+| `LANSENGER_APP_ID` | App ID du robot personnel | `your-appid` |
+| `LANSENGER_APP_SECRET` | App Secret du robot personnel | `ABCDEF123456...` |
 | `LANSENGER_API_GATEWAY_URL` | URL de la passerelle API Lansenger (remplacement) | `https://open.e.lanxin.cn/open/apigw` |
 
 Les identifiants peuvent aussi être fournis via la configuration `openclaw.json` (voir Configuration optionnelle ci-dessous). Les valeurs de configuration sont prioritaires ; les variables d'environnement sont utilisées comme repli lorsque la configuration n'est pas définie.
@@ -165,8 +165,8 @@ Les identifiants peuvent aussi être fournis via la configuration `openclaw.json
 
 | Champ | Description | Valeur par défaut |
 |-------|-------------|-------------------|
-| `appId` | App ID du bot personnel | — |
-| `appSecret` | App Secret du bot personnel | — |
+| `appId` | App ID du robot personnel | — |
+| `appSecret` | App Secret du robot personnel | — |
 | `apiGatewayUrl` | URL de la passerelle API | `https://open.e.lanxin.cn/open/apigw` |
 | `homeChannel` | Chat ID par défaut pour la livraison cron/notification | — |
 | `enabled` | Activer/désactiver le canal (défaut runtime : false sans identifiants) | `true` |
@@ -174,9 +174,9 @@ Les identifiants peuvent aussi être fournis via la configuration `openclaw.json
 | `dmPolicy` | Politique DM : `pairing`, `allowlist`, `open`, `disabled` | `pairing` |
 | `configWrites` | Autoriser Lansenger à écrire la config en réponse aux événements du canal | `true` |
 | `name` | Nom d'affichage pour ce compte | — |
-| `accounts` | Configuration multi-bot | — |
+| `accounts` | Configuration multi-robot | — |
 | `groupPolicy` | Politique de groupe : `open` (tous les groupes), `allowlist` (groupes autorisés uniquement), `disabled` (messages de groupe désactivés) | `allowlist` |
-| `groupAllowFrom` | IDs de groupes autorisés à déclencher le bot | `[]` |
+| `groupAllowFrom` | IDs de groupes autorisés à déclencher le robot | `[]` |
 | `groups` | Configuration par groupe (requireMention, enabled, allowFrom) | — |
 | `ackMessage` | Envoyer un message de confirmation avant le traitement de l'agent | `true` |
 | `revokeAckMessage` | Révoquer automatiquement le message ack après la réponse de l'agent. Mettre `false` pour garder le message visible (certains utilisateurs préfèrent voir le message plutôt qu'une notification « message révoqué ») | `false` |
@@ -208,12 +208,12 @@ Lorsqu'un utilisateur envoie plusieurs messages rapides consécutifs, le mécani
 - Les messages média et commandes de contrôle ne sont PAS debounceés — ils sont traités immédiatement
 - Quand le debounce est actif, les textes fusionnés sont joints par `\n` ; les chemins média sont concaténés ; les métadonnées du dernier message sont utilisées
 
-### Configuration multi-bot
+### Configuration multi-robot
 
-Pour ajouter plusieurs bots, utilisez `openclaw config set` avec la structure `accounts` :
+Pour ajouter plusieurs robots, utilisez `openclaw config set` avec la structure `accounts` :
 
 ```bash
-# Ajouter un deuxième bot (remplacez appid/appsecret/gateway par vos valeurs)
+# Ajouter un deuxième robot (remplacez appid/appsecret/gateway par vos valeurs)
 openclaw config set channels.lansenger.accounts.your-appid-2.appId "your-appid-2"
 openclaw config set channels.lansenger.accounts.your-appid-2.appSecret "your-appsecret"
 openclaw config set channels.lansenger.accounts.your-appid-2.apiGatewayUrl "https://apigw.lx.qianxin.com"
@@ -348,7 +348,7 @@ Le plugin supporte les cartes d'approbation :
 ## Notes importantes
 
 - **Pas de chat staff** — Lansenger n'a que les chats de groupe et DM (privé) ; il n'existe pas de concept de « chat staff ».
-- **Révocation chatType** — uniquement `bot` ou `group` ; pas de chatType `staff`.
+- **Révocation chatType** — uniquement `robot` ou `group` ; pas de chatType `staff`.
 - **Pas de sysMsg sur révocation** — l'API accepte `sysMsg` mais ne l'affiche pas.
 - **Pas de deleteMessage** — l'API retourne l'erreur 10000 ; la suppression n'est pas disponible.
 - **appArticles** — utilise le champ `summary` (pas `description`).
@@ -432,7 +432,7 @@ Voir [Documentation des appareils OpenClaw](https://docs.openclaw.ai/cli/devices
 
 ### "Le client mobile ne permet pas de voir les identifiants"
 
-Utilisez uniquement le **client Lansenger (desktop)**. L'application mobile n'affiche pas les identifiants du bot.
+Utilisez uniquement le **client Lansenger (desktop)**. L'application mobile n'affiche pas les identifiants du robot.
 
 ### "No binding for botId"
 

package/README.zhHant.md

@@ -22,7 +22,7 @@
 - **入站媒體處理**——下載圖片/檔案/語音，偵測副檔名，向代理提供檔案路徑
 - **訊息撤回**——撤回已傳送的訊息（chatType 僅支援 bot 與 group）
 - **自動啟動**——閘道啟動時自動連接所有已設定的機器人帳戶
-- **入站防抖合併**——利用 OpenClaw 的 `messages.inbound.debounceMs` 配置，合併同一發送者的連續快速訊息
+- **入站防抖合併**——利用 OpenClaw 的 `messages.inbound.debounceMs` 配置，合併同一傳送者的連續快速訊息
 - **確認訊息**——在代理處理前傳送「收到，正在處理...」確認訊息，代理回覆後自動撤回，語言自動偵測
 - **零核心修改**——純插件模式，`git diff HEAD` 保持 PRISTINE
 
@@ -50,13 +50,13 @@
 
 | 工具 | 說明 |
 |------|------|
-| `lansenger_send_text` | 發送純文字訊息，不支援 Markdown |
-| `lansenger_send_format_text` | 發送 Markdown 格式文字，支援 @提及 |
-| `lansenger_send_file` | 發送檔案/圖片/影片/語音（工作區或外部路徑） |
-| `lansenger_send_image_url` | 透過 URL 發送圖片 |
-| `lansenger_send_link_card` | 發送富連結預覽卡片 |
-| `lansenger_send_app_card` | 發送互動/審批卡片 |
-| `lansenger_send_app_articles` | 發送多文章卡片 |
+| `lansenger_send_text` | 傳送純文字訊息，不支援 Markdown |
+| `lansenger_send_format_text` | 傳送 Markdown 格式文字，支援 @提及 |
+| `lansenger_send_file` | 傳送檔案/圖片/影片/語音（工作區或外部路徑） |
+| `lansenger_send_image_url` | 透過 URL 傳送圖片 |
+| `lansenger_send_link_card` | 傳送富連結預覽卡片 |
+| `lansenger_send_app_card` | 傳送互動/審批卡片 |
+| `lansenger_send_app_articles` | 傳送多文章卡片 |
 | `lansenger_update_dynamic_card` | 原地更新動態卡片狀態 |
 | `lansenger_revoke_message` | 撤回已傳送的訊息 |
 | `lansenger_query_groups` | 查詢可用群組 |
@@ -201,7 +201,7 @@ openclaw pairing approve lansenger <配對碼>
 
 | 欄位 | 說明 | 預設值 |
 |------|------|--------|
-| `messages.inbound.debounceMs` | 全域防抖視窗（毫秒）；同一發送者在視窗內的連續訊息會被合併 | `0`（停用） |
+| `messages.inbound.debounceMs` | 全域防抖視窗（毫秒）；同一傳送者在視窗內的連續訊息會被合併 | `0`（停用） |
 | `messages.inbound.byChannel.lansenger` | 藍信頻道專屬覆蓋（優先於全域） | — |
 | `messages.queue.mode` | 代理處理中的佇列模式：`steer`、`followup`、`collect`、`queue`、`interrupt` | `steer`（推薦） |
 

package/README.zhHantHK.md

@@ -14,7 +14,7 @@
 - **即時訊息** — 透過 WebSocket 長連線實現
 - **多機械人支援** — 將多個藍信機械人綁定至不同的 OpenClaw 代理
 - **Markdown 支援** — 使用 `formatText` msgType（預設）
-- **檔案/圖片/語音附件** — 透過 `text` msgType 上傳媒體
+- **檔案/圖片/語音附件** — 透過 `text` msgType 上載媒體
 - **審批卡片**——互動式審批流程，支援原地狀態更新（待審批 → 已通過/已拒絕）
 - **語言偵測** — 自動偵測使用者語言，提供本地化回應
 - **msgTarget 自動路由** — 所有發送方法自動路由至群組聊天或私聊（DM）API；無需分別的群組/私聊方法
@@ -83,7 +83,7 @@ openclaw gateway restart
 
 > **可選**：安裝 `lansenger-cli` 作為 CLI 替代方案：`pipx install lansenger-cli`。
 
-> **自訂閘道**：企業私有化部署（如奇安信）需在設定後透過 `openclaw.json` 或環境變數設定 `apiGatewayUrl` — 見[可選設定](#可選設定)。
+> **自訂網關**：企業私有化部署（如奇安信）需在設定後透過 `openclaw.json` 或環境變數設定 `apiGatewayUrl` — 見[可選設定](#可選設定)。
 
 ### 開發安裝（本地連結）
 
@@ -180,7 +180,7 @@ openclaw pairing approve lansenger <配對碼>
 | `groups` | 羣級設定（requireMention、enabled、allowFrom） | — |
 | `ackMessage` | 在代理處理前發送確認訊息 | `true` |
 | `revokeAckMessage` | 代理回覆遞送後自動撤回確認訊息。設為 `false` 則保留確認訊息可見（有些使用者偏好看到確認訊息而非撤回的系統通知） | `false` |
-| `mediaLocalRoots` | 透過媒體傳送本地檔案時允許的根目錄列表；空陣列預設為工作區 + `/tmp` | `[cwd, /tmp]` |
+| `mediaLocalRoots` | 透過媒體發送本地檔案時允許的根目錄列表；空陣列預設為工作區 + `/tmp` | `[cwd, /tmp]` |
 | `ackMessageTextZh` | 中文確認訊息文案 | `收到，正在處理...` |
 | `ackMessageTextEn` | 英文確認訊息文案 | `Received, processing...` |
 

package/dist/src/channel.js

@@ -8,59 +8,20 @@ import { coerceSecretRef } from "openclaw/plugin-sdk/secret-ref-runtime";
 import { isPrivateNetworkOptInEnabled } from "openclaw/plugin-sdk/ssrf-policy";
 import { chunkMarkdownTextWithMode, resolveTextChunkLimit } from "openclaw/plugin-sdk/reply-chunking";
 import * as fs from "node:fs/promises";
-import * as fsSync from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
 import * as crypto from "node:crypto";
 import { LansengerClient, DEFAULT_API_GATEWAY_URL } from "./client.js";
 import { createMessageReceiptFromOutboundResults, createChannelMessageAdapterFromOutbound, } from "openclaw/plugin-sdk/channel-outbound";
-import { getRunningClient, getLastInboundTime, stripOpenClawUuidSuffix, gatewayStartAccount, gatewayStopAccount } from "./runtime.js";
+import { getRunningEntryByAccount, getLastInboundTimeByAccount, stripOpenClawUuidSuffix, gatewayStartAccount, gatewayStopAccount } from "./runtime.js";
 import { lansengerSetupWizard } from "./setup-wizard.js";
+import { PersistentStore } from "./persistent-store.js";
 const log = createSubsystemLogger("lansenger");
 const LANSENGER_TEXT_CHUNK_LIMIT = 4000;
 const APPROVAL_CARD_FILE = path.join(os.homedir(), ".openclaw", "lansenger-approval-cards.json");
-class PersistentApprovalCardStore {
-    data = new Map();
+class PersistentApprovalCardStore extends PersistentStore {
     constructor() {
-        this.load();
-    }
-    load() {
-        try {
-            const raw = fsSync.readFileSync(APPROVAL_CARD_FILE, "utf-8");
-            const parsed = JSON.parse(raw);
-            if (parsed && typeof parsed === "object") {
-                for (const [k, v] of Object.entries(parsed)) {
-                    this.data.set(k, v);
-                }
-            }
-        }
-        catch { }
-    }
-    save() {
-        try {
-            const dir = path.dirname(APPROVAL_CARD_FILE);
-            fsSync.mkdirSync(dir, { recursive: true });
-            const obj = {};
-            for (const [k, v] of this.data)
-                obj[k] = v;
-            fsSync.writeFileSync(APPROVAL_CARD_FILE, JSON.stringify(obj), "utf-8");
-        }
-        catch (e) {
-            log.warn(`failed to persist approval cards: ${e instanceof Error ? e.message : String(e)}`);
-        }
-    }
-    get(chatId) { return this.data.get(chatId); }
-    set(chatId, info) {
-        this.data.set(chatId, info);
-        this.save();
-    }
-    delete(chatId) {
-        this.data.delete(chatId);
-        this.save();
-    }
-    clear() {
-        this.data.clear();
-        this.save();
+        super(APPROVAL_CARD_FILE, "approval cards");
     }
 }
 const pendingApprovalCards = new PersistentApprovalCardStore();
@@ -456,14 +417,14 @@ const chatPlugin = createChatChannelPlugin({
             sendText: async (ctx) => {
                 const sessionKey = ctx.sessionKey;
                 const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-                const client = makeClient(account);
+                const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
                 const result = await client.sendFormatText(ctx.to, ctx.text);
                 return toOutboundResult(result.messageId, sessionKey, ctx.to);
             },
             sendMedia: async (ctx) => {
                 const sessionKey = ctx.sessionKey;
                 const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-                const client = makeClient(account);
+                const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
                 const caption = ctx.text ?? "";
                 if (ctx.mediaUrl && /^https?:\/\//i.test(ctx.mediaUrl)) {
                     const result = await client.sendImageUrl(ctx.to, ctx.mediaUrl, caption, account.dangerouslyAllowPrivateNetwork);
@@ -520,15 +481,17 @@ const chatPlugin = createChatChannelPlugin({
                 const sessionKey = params.sessionKey ?? "";
                 if (hint?.kind === "approval-resolved") {
                     const chatId = target?.to;
-                    const cardInfo = pendingApprovalCards.get(chatId);
+                    const acctId = target?.accountId ?? "";
+                    const cardKey = acctId ? `${acctId}:${chatId}` : chatId;
+                    const cardInfo = pendingApprovalCards.get(cardKey);
                     if (cardInfo?.messageId) {
                         const account = resolveAccount(params.cfg, target?.accountId ?? undefined);
-                        const client = makeClient(account);
+                        const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
                         const status = payload?.text?.includes("✅") ? "approved" : "denied";
                         try {
                             await client.updateCardStatus(cardInfo.messageId, status, cardInfo.lang);
                             log.info(`beforeDeliverPayload: approval card updated — messageId=${cardInfo.messageId} status=${status}`);
-                            pendingApprovalCards.delete(chatId);
+                            pendingApprovalCards.delete(cardKey);
                         }
                         catch (e) {
                             log.error(`beforeDeliverPayload: card update failed — ${e instanceof Error ? e.message : String(e)}`);
@@ -546,7 +509,7 @@ const chatPlugin = createChatChannelPlugin({
             sendFormattedText: async (ctx) => {
                 const sessionKey = ctx.sessionKey;
                 const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-                const client = makeClient(account);
+                const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
                 const result = await client.sendFormatText(ctx.to, ctx.text);
                 return [{ channel: "lansenger", messageId: result.messageId ?? "", sessionKey }];
             },
@@ -572,7 +535,7 @@ const chatPlugin = createChatChannelPlugin({
 const lansengerOutboundBridge = {
     sendText: async (ctx) => {
         const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-        const client = makeClient(account);
+        const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
         const result = await client.sendFormatText(ctx.to, ctx.text);
         const receipt = result.messageId
             ? createMessageReceiptFromOutboundResults({ results: [{ messageId: result.messageId, chatId: ctx.to }], kind: "text", sentAt: Date.now() })
@@ -581,7 +544,7 @@ const lansengerOutboundBridge = {
     },
     sendMedia: async (ctx) => {
         const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-        const client = makeClient(account);
+        const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
         const caption = ctx.text ?? "";
         if (ctx.mediaUrl && /^https?:\/\//i.test(ctx.mediaUrl)) {
             const result = await client.sendImageUrl(ctx.to, ctx.mediaUrl, caption, account.dangerouslyAllowPrivateNetwork);
@@ -630,7 +593,7 @@ const lansengerOutboundBridge = {
     },
     sendPayload: async (ctx) => {
         const account = resolveAccount(ctx.cfg, ctx.accountId ?? undefined);
-        const client = makeClient(account);
+        const client = getRunningEntryByAccount(account.accountId ?? "")?.client ?? makeClient(account);
         const payload = ctx.payload;
         const hasCodeBlock = payload?.text && /```/.test(payload.text);
         if (payload?.text && !payload?.mediaUrl && !payload?.presentation) {
@@ -901,9 +864,9 @@ export const lansengerPlugin = {
             return await probeLansengerAccount(account);
         },
         buildAccountSnapshot: ({ account, runtime, probe }) => {
-            const liveClient = getRunningClient();
-            const connected = liveClient?.isWsAlive() ?? false;
-            const lastInboundAt = getLastInboundTime();
+            const entry = getRunningEntryByAccount(account.accountId ?? "");
+            const connected = entry?.client.isWsAlive() ?? false;
+            const lastInboundAt = getLastInboundTimeByAccount(account.accountId ?? "");
             return {
                 accountId: account.accountId ?? account.appId ?? DEFAULT_ACCOUNT_ID,
                 enabled: account.enabled,
@@ -1016,7 +979,7 @@ export const lansengerPlugin = {
                         const appCard = (lang === "zh" ? payload.zh : payload.en) ?? payload.en;
                         const result = await client.sendAppCard(target.to, appCard);
                         if (result.messageId) {
-                            pendingApprovalCards.set(target.to, { messageId: result.messageId, lang: lang });
+                            pendingApprovalCards.set(`${accountId}:${target.to}`, { messageId: result.messageId, lang: lang });
                         }
                         return { delivered: result.success, messageId: result.messageId ?? null };
                     }