From 13cb750a2a703c98a6b4cee8a166c327c26c8102 Mon Sep 17 00:00:00 2001
From: Rogee <rogee@ipao.vip>
Date: Fri, 5 Jun 2026 13:38:55 +0800
Subject: [PATCH] feat(captain): align task payload persistence

---
 docs/CHATWOOT_PARITY_DEVELOPMENT_PLAN.md      |  41 ++-
 internal/app/bootstrap.go                     |   4 +-
 .../api/v1/captain_task_extended_handler.go   |  46 ++-
 .../v1/captain_task_extended_handler_test.go  |  78 ++++-
 .../handler/api/v1/captain_task_handler.go    |  29 +-
 .../api/v1/captain_task_handler_test.go       | 103 +++++++
 internal/model/copilot_models.go              |  27 ++
 .../repository/copilot_message_repo_test.go   |  43 ++-
 .../service/captain_task_extended_service.go  | 207 +++++++++++++-
 internal/service/captain_task_service.go      | 267 ++++++++++++++----
 internal/service/captain_task_service_test.go |  66 ++++-
 11 files changed, 822 insertions(+), 89 deletions(-)
 create mode 100644 internal/handler/api/v1/captain_task_handler_test.go

diff --git a/docs/CHATWOOT_PARITY_DEVELOPMENT_PLAN.md b/docs/CHATWOOT_PARITY_DEVELOPMENT_PLAN.md
index 21f968d9..47ad3035 100644
--- a/docs/CHATWOOT_PARITY_DEVELOPMENT_PLAN.md
+++ b/docs/CHATWOOT_PARITY_DEVELOPMENT_PLAN.md
@@ -16,9 +16,9 @@ Build GoChat as a Go backend that can directly reuse the frontend from `referenc
 
 ## Current Baseline
 
-- Latest implementation checkpoint: `feat(captain): gate document sync backend`.
-- Latest documentation checkpoint: this checkpoint, recorded with the B11.3d/B11.3e/B12 executable landing plan.
-- Worktree status at this documentation checkpoint: B11.1a aligns Captain assistant CRUD/tools/inbox bindings; B11.1b aligns Captain scenarios and custom tools; B11.1c aligns Captain documents, assistant responses, bulk actions, and custom-tool test payloads; B11.2 aligns Copilot thread/message create/list/get/delete payloads, account/user scoping, and no-LLM fallback persistence; B11.3a aligns Captain preferences show/update payloads and account-level model/feature storage; B11.3b aligns Captain playground request/response payloads, account scoping, v2 history handling, and no-LLM fallback; B11.3c adds the fakeable Captain document sync backend gate with disabled, failed, and fake-success states. Next active implementation slice is B11.3d Copilot tasks/tool-call depth, followed by B11.3e streaming/realtime compatibility and B12 reused frontend smoke.
+- Latest implementation checkpoint: this checkpoint, prepared as `feat(captain): align task payload persistence`.
+- Latest documentation checkpoint: this checkpoint, recorded with the B11.3d Captain task/tool-call persistence implementation.
+- Worktree status at this implementation checkpoint: B11.1a aligns Captain assistant CRUD/tools/inbox bindings; B11.1b aligns Captain scenarios and custom tools; B11.1c aligns Captain documents, assistant responses, bulk actions, and custom-tool test payloads; B11.2 aligns Copilot thread/message create/list/get/delete payloads, account/user scoping, and no-LLM fallback persistence; B11.3a aligns Captain preferences show/update payloads and account-level model/feature storage; B11.3b aligns Captain playground request/response payloads, account scoping, v2 history handling, and no-LLM fallback; B11.3c adds the fakeable Captain document sync backend gate with disabled, failed, and fake-success states; B11.3d aligns Captain task request/response payloads, no-provider disabled states, follow-up context, suggestion persistence, and Copilot message tool-call key validation. Next active implementation slice is B11.3e streaming/realtime compatibility, followed by B12 reused frontend smoke.
 - `go test ./...` passes.
 - Route dump succeeds with `TOTAL: 830` after adding the Chatwoot-compatible applied-SLA index route.
 - Route parity artifacts now exist under `docs/parity/` and are generated by `cmd/route_parity`.
@@ -44,9 +44,8 @@ Next ordered checkpoints:
 
 | Order | Slice | Required outcome | Primary verification |
 | --- | --- | --- | --- |
-| 1 | B11.3d | Copilot tasks, suggestions, and tool-call records persist enough frontend state for reloads. | Copilot task/tool-call handler/service fixtures and serializer tests. |
-| 2 | B11.3e | Streaming/realtime paths have frontend-compatible non-streaming or disabled fallbacks until true streaming is wired. | Deterministic response tests and event/push payload tests. |
-| 3 | B12 | Reused Chatwoot frontend smoke runs repeatably against GoChat. | Checked smoke command plus gap report under `docs/parity/`. |
+| 1 | B11.3e | Streaming/realtime paths have frontend-compatible non-streaming or disabled fallbacks until true streaming is wired. | Deterministic response tests and event/push payload tests. |
+| 2 | B12 | Reused Chatwoot frontend smoke runs repeatably against GoChat. | Checked smoke command plus gap report under `docs/parity/`. |
 
 ## Execution Snapshot
 
@@ -56,7 +55,7 @@ Next ordered checkpoints:
 | Phase 1 | Meilisearch search engine | Review | B6 payload parity, optional live gate, and DB-fallback hardening are implemented; an actual live Meilisearch run is optional and environment-dependent |
 | Phase 2 | Route and controller parity audit | Doing | Ruby/Bundler unavailable, so Chatwoot route extraction currently uses static `routes.rb` fallback |
 | Phase 3 | Data and serializer parity | Doing | JSON fixture coverage is partial and still endpoint-family based |
-| Phase 4 | Enterprise feature completion | Doing | B7, B8, B9, and B10 are in Review; B11 Captain resources, preferences, playground, document sync gates, and Copilot thread/message payloads are landed, while Copilot tasks/tool-call depth and streaming remain active |
+| Phase 4 | Enterprise feature completion | Doing | B7, B8, B9, and B10 are in Review; B11 Captain resources, preferences, playground, document sync gates, Copilot thread/message payloads, and Captain task payload persistence are landed, while streaming/realtime compatibility remains active |
 | Phase 5 | Background jobs and integrations | Planned | durable worker choice and job parity are open |
 | Phase 6 | Core placeholder burn-down | Doing | account/contact/conversation/message/inbox placeholder groups remain broad |
 | Phase 7 | Verification harness | Planned | search live gate and reused-frontend smoke harness are not complete |
@@ -156,6 +155,7 @@ This ledger records the committed parity checkpoints that future slices should b
 | `feat(captain): align playground fallback` | Completes B11.3b Captain assistant playground parity: `POST /captain/assistants/:assistant_id/playground` now accepts the Chatwoot frontend top-level `message_content` and `message_history` payload, enforces account-scoped assistant lookup, returns raw legacy `{ content }` or v2 `{ response }` payloads, appends the current user message once for `captain_integration_v2`, preserves history ordering through the LLM boundary, and returns a deterministic raw disabled response when no LLM provider is configured. | `go test ./internal/handler/api/v1 -run CaptainAssistant -count=1`; `go test ./internal/service -run Captain -count=1`; `go test ./internal/handler/api/v1 -run 'Captain\|Copilot' -count=1`; `go test ./internal/service -run 'Captain\|Copilot' -count=1`; `go test ./internal/handler/api/v1 -count=1`; `go test ./internal/service -count=1`; `go test ./...`; `git diff --check`. Focused handler fixtures cover raw no-envelope fallback, no-history default, cross-account rejection, v2 append/no-duplicate history behavior, and fake LLM request capture. | Continue B11.3c with Captain document sync/indexing and Meilisearch/embedding gates. |
 | `feat(captain): gate document sync backend` | Completes B11.3c's first document sync gate: Captain document sync now has a fakeable `CaptainDocumentSyncBackend` boundary and account-scoped `SyncDocumentByAccount` worker entry that marks disabled config as `sync_disabled`, records backend errors as `sync_error`, preserves backend error codes such as `access_denied`, stores successful content/title/fingerprint updates as `synced`, and avoids external HTTP/LLM calls in default tests. | `go test ./internal/service -run CaptainDocument -count=1`; `go test ./internal/service -run 'Captain\|Copilot' -count=1`; `go test ./internal/handler/api/v1 -run 'Captain\|Copilot' -count=1`; `go test ./internal/service -count=1`; `go test ./internal/handler/api/v1 -count=1`; `go test ./...`; `git diff --check`. Focused service fixtures cover disabled, fake success, backend error, backend error-code, content fingerprint normalization, and account scoping. | Continue B11.3d with Copilot task/tool-call persistence and suggestion payload depth. |
 | `docs: land copilot task execution plan` | Documentation-only checkpoint that fully lands the B11.3d/B11.3e/B12 execution plan: Captain task routes now have exact Chatwoot request/response contracts, current Go gaps, task persistence rules, follow-up context behavior, no-provider disabled-state rules, focused verification commands, streaming/realtime fallback gates, and reused frontend smoke deliverables. | `git diff --check`; no Go code changed. | Start B11.3d implementation from the recorded task/tool-call contract before moving to streaming or frontend smoke. |
+| `feat(captain): align task payload persistence` | Completes B11.3d's frontend task contract: rewrite/summarize/reply suggestion now accept Chatwoot `operation` and `conversation_display_id` payloads, return raw `{ message, follow_up_context }` or `422 { error }`, avoid no-provider panics with `Captain is disabled`, resolve conversations by account-scoped display ID, and persist task outputs into `copilot_suggestion_messages`. Label suggestion and follow-up POST routes now consume the dashboard `tasks.js` bodies, preserve/update Chatwoot follow-up context, and return raw payloads. `CopilotMessage` now validates the Chatwoot allowed JSON keys `content`, `reasoning`, `function_name`, and `reply_suggestion` for reloadable tool-call state. | `go test ./internal/handler/api/v1 -run 'CaptainTask\|Captain\|Copilot' -count=1`; `go test ./internal/service -run 'CaptainTask\|Captain\|Copilot' -count=1`; `go test ./internal/repository -run 'Copilot\|Captain' -count=1`; `go test ./internal/handler/api/v1 -count=1`; `go test ./internal/service -count=1`; `TMPDIR=/home/rogee/Projects/gochat/.tmp/test-tmp go test ./...` with existing caches; `git diff --check`. Focused fixtures cover raw no-envelope task responses, disabled state, display-ID lookup, suggestion persistence, follow-up history update, and Copilot message key validation. | Continue B11.3e streaming/realtime compatibility; B12 frontend smoke remains after B11.3e. |
 
 ## Next Slice Contract
 
@@ -163,7 +163,7 @@ Completed implementation slice: B11.3c now adds Captain document sync backend ga
 
 Completed documentation slice: the remaining development plan is now fully landed in this tracker. B11.3 is split into concrete Captain playground, document sync/indexing, Copilot task/tool-call, and streaming/realtime checkpoints. B12 has a repeatable smoke-report contract. B9.3 delayed/durable worker scheduling and B9.4 macro attachment depth remain named Phase 5 follow-ups.
 
-Next implementation slice: start B11.3d Copilot task/tool-call persistence and suggestion payload depth. Do not expand to streaming until rewrite/summarize/reply suggestion/label suggestion/follow-up request payloads, raw disabled states, and reloadable task/tool-call serializers are covered by tests.
+Next implementation slice: start B11.3e streaming/realtime compatibility. B11.3d now covers rewrite/summarize/reply suggestion/label suggestion/follow-up request payloads, raw disabled states, persisted suggestion state, and reloadable Copilot message key validation.
 
 | Step | Required result | Reference source | Verification |
 | --- | --- | --- | --- |
@@ -190,7 +190,7 @@ Next implementation slice: start B11.3d Copilot task/tool-call persistence and s
 | N21 | Keep B11.3a Captain preferences as current account-level Captain config baseline. | `preferences_controller.rb`, `CaptainFeaturable`, `config/llm.yml`, dashboard `preferences.js` API/store. | Done by `feat(captain): align preferences payloads`; raw model/provider/feature payloads, merge update behavior, account storage, admin gate, and exact preference routes are covered. |
 | N22 | Keep B11.3b Captain assistant playground payload parity as current playground baseline. | `reference/chatwoot/enterprise/app/controllers/api/v1/accounts/captain/assistants_controller.rb#playground`, dashboard `api/captain/assistant.js`, assistant playground component/store. | Done by `feat(captain): align playground fallback`; top-level `message_content` and `message_history` are accepted, account-scoped assistant lookup is enforced, legacy `{ content }` and v2 `{ response }` shapes are covered, and no-LLM fallback returns raw frontend-compatible JSON without 500. |
 | N23 | Keep B11.3c Captain document sync/indexing gates as current document-sync baseline. | Captain document sync service/jobs, document controller `sync`, existing Meilisearch engine, and any local embedding boundary. | Done by `feat(captain): gate document sync backend`; disabled config, fake successful indexing, failed sync metadata, fingerprint normalization, and account-scoped document lookup are covered without opening external network connections in default tests. |
-| N24 | Implement B11.3d Copilot task/tool-call persistence. | `resource :tasks` routes, Copilot/Captain task services, dashboard Copilot clients, current local `copilot_*` models. | Rewrite/summarize/reply suggestion/label suggestion/follow-up requests persist task/tool-call state or return documented raw disabled states; serializers survive frontend reload. |
+| N24 | Keep B11.3d Copilot task/tool-call persistence as current task baseline. | `resource :tasks` routes, Copilot/Captain task services, dashboard Copilot clients, current local `copilot_*` models. | Done by `feat(captain): align task payload persistence`; task routes accept frontend bodies, persist suggestions, return raw payloads/disabled states, and validate reloadable Copilot message keys. |
 | N25 | Implement B11.3e streaming/realtime compatibility. | Chatwoot Copilot/Captain streaming, push/event payloads, current Go channel dispatcher. | If streaming is not implemented, the endpoint returns deterministic non-streaming payloads and records the fallback; no route returns placeholder success or local envelope. |
 | N26 | Start B12 smoke harness after B11.3 has a tested disabled/external-provider story. | `reference/chatwoot` frontend boot scripts and GoChat dev/test boot flow. | A checked command and `docs/parity/frontend_smoke_report.md` record core and enterprise smoke status. |
 | N27 | Keep Hermes source plans mapped but inactive. | `.hermes/plans/2025-05-24-global-search-meilisearch.md`, `.hermes/plans/2026-05-24-automation-macro-csat.md`. | New work must update this tracker directly; Hermes files are read-only source notes unless the user asks otherwise. |
@@ -789,7 +789,7 @@ B11 Captain/Copilot breakdown:
 | B11.3a | Align Captain preferences show/update payloads, account-level model/feature storage, and admin update gate. | `preferences_controller.rb`, `CaptainFeaturable`, `config/llm.yml`, dashboard `preferences.js` API/store. | Handler tests cover raw `{ providers, models, features }`, default selected models, merge update behavior, account persistence, invalid model rejection, admin gate, and exact preference routes. | Done by `feat(captain): align preferences payloads` |
 | B11.3b | Align Captain assistant playground request/response behavior and no-LLM fallback. | `assistants_controller.rb#playground`, `Captain::Llm::AssistantChatService`, `AgentRunnerService`, dashboard `api/captain/assistant.js`, playground component/store. | Handler/service tests cover top-level `message_content`, `message_history`, account-scoped assistant lookup, no-history payloads, no duplicate latest user message for v2 behavior if enabled, raw `{ content }` or `{ response }` response shape, and safe no-LLM fallback. | Done by `feat(captain): align playground fallback` |
 | B11.3c | Implement document sync/indexing gates for Meilisearch or the chosen embedding/search backend. | Captain document sync/indexing services, document controller `sync`, existing Meilisearch engine, local embedding/config boundaries. | Tests cover disabled config, failed sync observability, successful fake backend indexing, account-scoped document lookup, and no external network call in default test mode. | Done by `feat(captain): gate document sync backend` |
-| B11.3d | Finish Copilot task, suggestion, and tool-call persistence depth. | `resource :tasks` routes in `reference/chatwoot/config/routes.rb:62-89`, Copilot/Captain task services, dashboard Copilot clients. | Tests cover rewrite/summarize/reply suggestion/label suggestion/follow-up request payloads, persisted task/tool-call records, reloadable serializers, account/user scoping, and raw disabled state when provider config is missing. | Todo |
+| B11.3d | Finish Copilot task, suggestion, and tool-call persistence depth. | `resource :tasks` routes in `reference/chatwoot/config/routes.rb:62-89`, Copilot/Captain task services, dashboard Copilot clients. | Tests cover rewrite/summarize/reply suggestion/label suggestion/follow-up request payloads, persisted task/tool-call records, reloadable serializers, account/user scoping, and raw disabled state when provider config is missing. | Done by `feat(captain): align task payload persistence` |
 | B11.3e | Add streaming/realtime compatibility or deterministic non-streaming fallback. | Chatwoot Captain/Copilot streaming paths, push payloads, current Go dispatcher/realtime boundaries. | Tests prove endpoints never return placeholder success/local envelopes; non-streaming fallback payloads are deterministic and frontend-compatible until true streaming is implemented. | Todo |
 
 B11.1a current checkpoint:
@@ -863,7 +863,7 @@ B11.3c-B11.3e follow-up contracts:
 | Slice | Landing rule | Status gate |
 | --- | --- | --- |
 | B11.3c document sync/indexing | Document sync must expose frontend-visible `syncing/synced/failed` states and call a fakeable Meilisearch/embedding boundary. Missing provider config must be a raw disabled/failure state, not a placeholder success. | B11 remains `Doing` until disabled, failed, and fake-success sync tests are present. |
-| B11.3d Copilot tasks/tool-calls | Rewrite, summarize, reply suggestion, label suggestion, and follow-up task routes must accept Chatwoot request payloads, persist enough task/tool-call output for reloads, and serialize raw frontend payloads. | B11 remains `Doing` until task/tool-call fixtures cover account/user scoping and no-provider fallback. |
+| B11.3d Copilot tasks/tool-calls | Rewrite, summarize, reply suggestion, label suggestion, and follow-up task routes must accept Chatwoot request payloads, persist enough task/tool-call output for reloads, and serialize raw frontend payloads. | Done by `feat(captain): align task payload persistence`; B11 remains `Doing` for B11.3e streaming/realtime fallback. |
 | B11.3e streaming/realtime | Streaming can be deferred only with a deterministic non-streaming response or explicit disabled state. Realtime push payloads should use the same thread/message serializer already landed in B11.2. | B11 moves to `Review` only after every streaming/realtime endpoint has a tested frontend-compatible fallback. |
 
 B11.3c current checkpoint:
@@ -906,6 +906,18 @@ env GOCACHE=/tmp/gochat-gocache GOMODCACHE=/tmp/gochat-gomodcache go test ./...
 git diff --check
 ```
 
+B11.3d current checkpoint:
+
+- Reference files inspected for this slice: `reference/chatwoot/app/controllers/api/v1/accounts/captain/tasks_controller.rb`, dashboard `api/captain/tasks.js`, `useCaptain.js`, `Captain::BaseTaskService`, `RewriteService`, `SummaryService`, `ReplySuggestionService`, `LabelSuggestionService`, `FollowUpService`, and enterprise `CopilotMessage`.
+- `rewrite`, `summarize`, and `reply_suggestion` now accept Chatwoot frontend `conversation_display_id` payloads, keep legacy local IDs only as service compatibility, and resolve conversations through account-scoped display-ID lookup.
+- Task handlers now return raw Chatwoot `{ message, follow_up_context }` success payloads and `422 { error }` disabled/error payloads instead of the local `response.OK` envelope.
+- Missing LLM provider returns the deterministic raw `Captain is disabled` state for task routes instead of panicking or producing `500`.
+- Rewrite validates Chatwoot operations: `fix_spelling_grammar`, `improve`, `casual`, `professional`, `friendly`, `confident`, and `straightforward`; invalid operations return `422`.
+- Label suggestion and follow-up POST routes now consume the reused dashboard JSON bodies. Follow-up validates allowed event names, appends user/assistant turns to `conversation_history`, updates `last_response`, and preserves `channel_type`.
+- Task outputs persist to `copilot_suggestion_messages` by account and resolved conversation so reloadable frontend state survives beyond the request.
+- `CopilotMessage` now rejects JSON keys outside Chatwoot's allowed set: `content`, `reasoning`, `function_name`, and `reply_suggestion`.
+- Focused handler/service/repository fixtures cover raw no-envelope payloads, display-ID lookup, no-provider disabled state, suggestion persistence, follow-up history updates, and Copilot message key validation.
+
 B11.3e streaming/realtime landing contract:
 
 | Area | Landing requirement | Done signal |
@@ -1064,7 +1076,7 @@ Remaining slice landing plan:
 | B8 | Done: account-side CSAT list/metrics/review-note payload parity, public submit/update idempotency, resolve-triggered generic survey message send, and CSV download filters/columns. | Channel-specific WhatsApp/Twilio template send hooks and durable queueing are tracked under Phase 5. | Review after `b36cf07`; move to Done only after channel-specific send hooks are either implemented or formally split as external integration scope. |
 | B9 | Done: B9.1 automation rule CRUD/listener/log/external-action parity and B9.2 macro frontend CRUD/execute side effects. | Delayed actions, durable queued worker scheduling, and deeper macro attachment/file parity remain named B9.3/B9.4 follow-ups. | Review after `feat(macros): align chatwoot macro payloads`; move to Done only after durable worker/attachment gaps are implemented or formally split out. |
 | B10 | Done: audit list payload, audit writer boundary for representative mutating core resources, CustomRole permission-key parity, AccountUser permission resolution, admin gates, delete nullification, and account-level InboxLimit enforcement in inbox/channel creation paths. | Frontend smoke coverage for enterprise settings remains B12, not a hidden B10 blocker. | Review after B10.4; move to Done only after reused frontend smoke confirms audit/custom-role/limit settings flows or any smoke gaps are split into owned follow-ups. |
-| B11 | Done through B11.3a: Captain Assistant CRUD, inbox binding, scenarios, documents, responses, custom tools, Copilot threads/messages, and Captain preferences payload fixtures. | B11.3b playground, B11.3c document sync/embedding gates, B11.3d Copilot tasks/tool-calls, and B11.3e streaming/realtime fallback. | Done only after LLM-dependent behavior is either implemented behind config or safely stubbed with frontend-compatible disabled states. |
+| B11 | Done through B11.3d: Captain Assistant CRUD, inbox binding, scenarios, documents, responses, custom tools, Copilot threads/messages, Captain preferences, playground, document sync gates, and task/tool-call payload persistence fixtures. | B11.3e streaming/realtime fallback remains active. | Done only after LLM-dependent behavior is either implemented behind config or safely stubbed with frontend-compatible disabled states. |
 | B12 | Boot reused Chatwoot frontend against GoChat auth/profile/inbox/conversation/contact flows. | Add smoke paths for widget init/message, public CSAT, reports, and enterprise screens as B7-B11 land. | Done only after the smoke command is repeatable and writes a checked gap report. |
 
 Per-slice documentation rule:
@@ -1342,7 +1354,7 @@ Enterprise tracking table:
 | P4.3 | Custom roles/permissions | `internal/model/custom_role.go`, `internal/service/custom_role_service.go`, `internal/middleware/role_check.go`, `internal/handler/api/v1/custom_role_handler.go` | Permission arrays, raw admin payloads, administrator gates, account-user resolution, profile permissions, and delete nullification are implemented; keep in Review until frontend smoke covers enterprise settings. | Review |
 | P4.4 | Agent capacity | `internal/model/agent_capacity_policy.go`, `internal/service/agent_capacity_policy_service.go`, `internal/handler/api/v1/agent_capacity_handler.go`, `internal/autoassignment/*` | First align AgentCapacityPolicy API, serializers, nested users, and `InboxCapacityLimit` data contracts; then enforce capacity in manual/auto assignment. | Review |
 | P4.5 | Inbox limits | `internal/model/agent_capacity_policy.go` `InboxCapacityLimit`, plus legacy `internal/model/inbox_limit.go`, `internal/service/inbox_limit_service.go`, `internal/repository/inbox_limit_repo.go`, `internal/handler/api/v1/inbox_limit_handler.go` | Chatwoot enterprise nested `agent_capacity_policies/:policy_id/inbox_limits`, assignment-path enforcement, and account-level `usage_limits[:inboxes]` create-path enforcement are covered. | Review |
-| P4.6 | Captain/Copilot | `internal/model/captain_models.go`, `internal/model/copilot_models.go`, `internal/service/captain_*`, `internal/service/copilot_*`, `internal/handler/api/v1/captain_*`, `internal/handler/api/v1/copilot_handler.go` | Captain resource payloads through B11.1c, Copilot thread/message payloads through B11.2, Captain preferences through B11.3a, Captain playground through B11.3b, and document sync gates through B11.3c are aligned; complete B11.3d Copilot task/tool-call depth and B11.3e streaming compatibility. | Doing |
+| P4.6 | Captain/Copilot | `internal/model/captain_models.go`, `internal/model/copilot_models.go`, `internal/service/captain_*`, `internal/service/copilot_*`, `internal/handler/api/v1/captain_*`, `internal/handler/api/v1/copilot_handler.go` | Captain resource payloads through B11.1c, Copilot thread/message payloads through B11.2, Captain preferences through B11.3a, Captain playground through B11.3b, document sync gates through B11.3c, and task/tool-call payload persistence through B11.3d are aligned; complete B11.3e streaming compatibility. | Doing |
 | P4.7 | CSAT | `internal/csat/*`, `internal/automation/csat_survey_*`, `internal/handler/api/v1/csat_*`, `internal/service/csat_metrics_service.go` | Account-side list/metrics/review-note payloads, public update depth, resolve-triggered generic survey send, and download CSV are aligned; channel-specific template send hooks remain Phase 5 integration work. | Review |
 | P4.8 | Automation and macros | `internal/automation/*`, `internal/handler/api/v1/automation_rule_handler.go`, `internal/handler/api/v1/macro_handler.go` | Automation-rule CRUD/listener/log/external-action delivery and macro CRUD/display-ID execution parity are implemented; durable delayed actions and deeper macro attachment/file parity remain B9.3/B9.4 follow-ups. | Review |
 | P4.9 | Assignment policies | `internal/autoassignment/*`, `internal/automation/agent_bot_rule_listener.go` | Match Chatwoot assignment policy behavior and availability/capacity rules. | Doing |
@@ -1359,7 +1371,7 @@ Enterprise work package breakdown:
 | Audit | Audit model parity, mutating action coverage, request metadata, filters/pagination, admin endpoint payloads. | List/admin payload parity is covered by `feat(audit): align chatwoot audit log payloads`; named representative writer coverage is covered by `feat(audit): record enterprise mutations` and `feat(audit): cover operational mutations`. | Review |
 | Custom roles | Permission-key parity, account-user role resolution, policy middleware, create/update/delete behavior. | Permission arrays, raw admin payloads, admin gates, RBAC/account-scope/profile resolution, and delete nullification are covered by `feat(custom-roles): align chatwoot permissions`. | Review |
 | Inbox limits | Account/inbox limit models, create/update enforcement, UI-readable limit responses, admin overrides. | Chatwoot nested capacity-limit API, assignment-path enforcement, and account-level create-limit `402` behavior are covered by B5/B7/B10.4. | Review |
-| Captain/Copilot | Assistants, inbox bindings, scenarios, responses, documents, tools, preferences, copilot threads/messages, tasks, streaming/tool-call behavior. | Assistant CRUD/tools/inbox binding fixtures are covered by `feat(captain): align assistant resources`; scenario/custom-tool fixtures are covered by `feat(captain): align scenario and tool payloads`; document/assistant-response/bulk/custom-tool-test fixtures are covered by `feat(captain): align document response actions`; Copilot thread/message fixtures are covered by `feat(copilot): align thread message payloads`; preferences fixtures are covered by `feat(captain): align preferences payloads`; playground fixtures are covered by `feat(captain): align playground fallback`; document sync gate fixtures are covered by `feat(captain): gate document sync backend`; B11.3d-B11.3e now track remaining Copilot task/tool-call behavior, streaming fallback, and frontend smoke screens. | Doing |
+| Captain/Copilot | Assistants, inbox bindings, scenarios, responses, documents, tools, preferences, copilot threads/messages, tasks, streaming/tool-call behavior. | Assistant CRUD/tools/inbox binding fixtures are covered by `feat(captain): align assistant resources`; scenario/custom-tool fixtures are covered by `feat(captain): align scenario and tool payloads`; document/assistant-response/bulk/custom-tool-test fixtures are covered by `feat(captain): align document response actions`; Copilot thread/message fixtures are covered by `feat(copilot): align thread message payloads`; preferences fixtures are covered by `feat(captain): align preferences payloads`; playground fixtures are covered by `feat(captain): align playground fallback`; document sync gate fixtures are covered by `feat(captain): gate document sync backend`; task/tool-call payload persistence is covered by `feat(captain): align task payload persistence`; B11.3e now tracks remaining streaming fallback and frontend smoke screens. | Doing |
 
 Enterprise acceptance gates:
 
@@ -1602,3 +1614,4 @@ Verification milestone gates:
 - 2026-06-05: B11.3b Captain playground checkpoint prepared as `feat(captain): align playground fallback`; playground now accepts Chatwoot top-level `message_content/message_history`, scopes assistant lookup to the account, returns raw legacy `{ content }` or v2 `{ response }`, appends current v2 user messages without duplication, preserves history into the fakeable LLM boundary, and returns deterministic no-LLM fallback JSON instead of 500. Focused CaptainAssistant/Captain tests, handler/service package tests, full `go test ./...`, and `git diff --check` passed. Next slice is B11.3c document sync/indexing and Meilisearch/embedding gates.
 - 2026-06-05: B11.3c Captain document sync gate checkpoint prepared as `feat(captain): gate document sync backend`; document sync now has a fakeable backend boundary and account-scoped worker entry, disabled config is recorded as `sync_disabled`, backend failures become `sync_error` or preserve explicit backend error codes, fake success updates content/title/fingerprint/timestamps as `synced`, and default tests avoid external network/LLM calls. Focused Captain/Copilot tests, handler/service package tests, full `go test ./...`, and `git diff --check` passed. Next slice is B11.3d Copilot task/tool-call persistence.
 - 2026-06-05: B11.3d/B11.3e/B12 planning checkpoint prepared as `docs: land copilot task execution plan`; the active tracker now records exact Chatwoot Captain task request/response contracts, current Go handler/service gaps, follow-up context and allowed operation rules, reloadable Copilot message/tool-call persistence requirements, no-provider disabled-state expectations, streaming/realtime fallback gates, B12 smoke harness deliverables, and focused exit commands. Documentation-only checkpoint; `git diff --check` passed. Next slice is B11.3d implementation.
+- 2026-06-05: B11.3d Captain task/tool-call checkpoint prepared as `feat(captain): align task payload persistence`; rewrite/summarize/reply suggestion now accept Chatwoot task payloads and return raw `{ message, follow_up_context }` or `422 { error }`, label suggestion and follow-up POST routes consume dashboard `tasks.js` bodies, no-provider paths return `Captain is disabled`, task outputs persist to `copilot_suggestion_messages`, and `CopilotMessage` validates reloadable tool-call keys. Focused CaptainTask/Captain/Copilot handler and service tests, Copilot/Captain repository tests, handler/service package tests, full `go test ./...` with workspace `TMPDIR`, and `git diff --check` passed. Next slice is B11.3e streaming/realtime compatibility.
diff --git a/internal/app/bootstrap.go b/internal/app/bootstrap.go
index 0db6d84e..80e7a8e9 100644
--- a/internal/app/bootstrap.go
+++ b/internal/app/bootstrap.go
@@ -565,10 +565,10 @@ func Bootstrap(env string) (*App, error) {
 	captainCustomToolService := service.NewCaptainCustomToolService(captainCustomToolRepo)
 	copilotService := service.NewCopilotService(copilotThreadRepo, copilotMessageRepo, copilotSuggestionRepo, llmProvider, captainAssistantRepo)
 	copilotContextService := service.NewCopilotContextService(messageRepo, conversationRepo, contactRepo, llmProvider)
-	captainTaskService := service.NewCaptainTaskService(captainAssistantRepo, captainAssistantResponseRepo, captainCustomToolRepo, conversationRepo, messageRepo, llmProvider, copilotContextService)
+	captainTaskService := service.NewCaptainTaskService(captainAssistantRepo, captainAssistantResponseRepo, captainCustomToolRepo, conversationRepo, messageRepo, llmProvider, copilotContextService, copilotSuggestionRepo)
 	conversationInsightService := service.NewConversationInsightService(conversationRepo, messageRepo, captainAssistantRepo, llmProvider)
 	captainPreferenceService := service.NewCaptainPreferenceService(captainPreferenceRepo, accountRepo)
-	captainTaskExtendedService := service.NewCaptainTaskExtendedService(conversationRepo, messageRepo, captainAssistantRepo, captainPreferenceRepo, llmProvider)
+	captainTaskExtendedService := service.NewCaptainTaskExtendedService(conversationRepo, messageRepo, captainAssistantRepo, captainPreferenceRepo, llmProvider, copilotSuggestionRepo)
 	captainAssistantResponseService := service.NewCaptainAssistantResponseService(captainAssistantRepo, captainAssistantResponseRepo, conversationRepo, messageRepo, captainPreferenceRepo, llmProvider)
 	captainBulkActionService := service.NewCaptainBulkActionService(conversationRepo, messageRepo, captainAssistantRepo, captainPreferenceRepo, llmProvider, captainTaskExtendedService, captainAssistantResponseService)
 	captainBulkActionService.SetCaptainResourceRepos(captainAssistantResponseRepo, captainDocumentRepo)
diff --git a/internal/handler/api/v1/captain_task_extended_handler.go b/internal/handler/api/v1/captain_task_extended_handler.go
index 4965126d..470379e3 100644
--- a/internal/handler/api/v1/captain_task_extended_handler.go
+++ b/internal/handler/api/v1/captain_task_extended_handler.go
@@ -30,6 +30,22 @@ func (h *CaptainTaskExtendedHandler) LabelSuggestion(c *gin.Context) {
 		return
 	}
 
+	if c.Request.Method == http.MethodPost {
+		var req service.ChatwootLabelSuggestionRequest
+		if err := c.ShouldBindJSON(&req); err != nil {
+			response.AbortWithStatusError(c, http.StatusBadRequest, response.ErrBadRequest, "invalid request body: "+err.Error())
+			return
+		}
+		result, err := h.svc.LabelSuggestion(c.Request.Context(), uint(accountID), &req)
+		if err != nil {
+			applogger.L().Errorf("Label suggestion: %v", err)
+			renderCaptainTaskError(c, err)
+			return
+		}
+		renderCaptainExtendedTaskResult(c, result)
+		return
+	}
+
 	// Parse conversation_ids from query param (comma-separated)
 	convIDsStr := c.Query("conversation_ids")
 	if convIDsStr == "" {
@@ -69,6 +85,22 @@ func (h *CaptainTaskExtendedHandler) FollowUp(c *gin.Context) {
 		return
 	}
 
+	if c.Request.Method == http.MethodPost {
+		var req service.ChatwootFollowUpRequest
+		if err := c.ShouldBindJSON(&req); err != nil {
+			response.AbortWithStatusError(c, http.StatusBadRequest, response.ErrBadRequest, "invalid request body: "+err.Error())
+			return
+		}
+		result, err := h.svc.FollowUp(c.Request.Context(), uint(accountID), &req)
+		if err != nil {
+			applogger.L().Errorf("Follow-up: %v", err)
+			renderCaptainTaskError(c, err)
+			return
+		}
+		renderCaptainExtendedTaskResult(c, result)
+		return
+	}
+
 	convIDsStr := c.Query("conversation_ids")
 	if convIDsStr == "" {
 		response.AbortWithStatusError(c, http.StatusBadRequest, response.ErrValidation, "conversation_ids required")
@@ -98,6 +130,18 @@ func (h *CaptainTaskExtendedHandler) FollowUp(c *gin.Context) {
 	response.OK(c, result)
 }
 
+func renderCaptainExtendedTaskResult(c *gin.Context, result *service.ChatwootTaskResult) {
+	if result == nil || result.Message == nil {
+		c.JSON(http.StatusOK, gin.H{"message": nil})
+		return
+	}
+	payload := gin.H{"message": *result.Message}
+	if result.FollowUpContext != nil {
+		payload["follow_up_context"] = result.FollowUpContext
+	}
+	c.JSON(http.StatusOK, payload)
+}
+
 // parseUintSlice parses a comma-separated string of uint values.
 func parseUintSlice(s string) ([]uint, error) {
 	parts := strings.Split(s, ",")
@@ -114,4 +158,4 @@ func parseUintSlice(s string) ([]uint, error) {
 		result = append(result, uint(v))
 	}
 	return result, nil
-}
\ No newline at end of file
+}
diff --git a/internal/handler/api/v1/captain_task_extended_handler_test.go b/internal/handler/api/v1/captain_task_extended_handler_test.go
index f3d04a3c..8f15e773 100644
--- a/internal/handler/api/v1/captain_task_extended_handler_test.go
+++ b/internal/handler/api/v1/captain_task_extended_handler_test.go
@@ -1,6 +1,7 @@
 package v1
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"fmt"
@@ -47,12 +48,14 @@ func setupTaskExtendedHandlerTest(t *testing.T, mockLLM llm.Provider) (*CaptainT
 		&model.Message{},
 		&model.Inbox{},
 		&model.Contact{},
+		&model.CopilotSuggestionMessage{},
 	))
 	convRepo := repository.NewConversationRepo(db)
 	msgRepo := repository.NewMessageRepo(db)
 	assistantRepo := repository.NewCaptainAssistantRepo(db)
 	prefRepo := repository.NewCaptainPreferenceRepo(db)
-	svc := service.NewCaptainTaskExtendedService(convRepo, msgRepo, assistantRepo, prefRepo, mockLLM)
+	suggestionRepo := repository.NewCopilotSuggestionRepo(db)
+	svc := service.NewCaptainTaskExtendedService(convRepo, msgRepo, assistantRepo, prefRepo, mockLLM, suggestionRepo)
 	handler := NewCaptainTaskExtendedHandler(svc)
 	return handler, db
 }
@@ -103,6 +106,40 @@ func TestCaptainTaskExtendedHandler_LabelSuggestion_MissingConversationIDs(t *te
 	assert.Equal(t, http.StatusBadRequest, w.Code)
 }
 
+func TestCaptainTaskExtendedHandler_LabelSuggestion_ChatwootPostRawPayload(t *testing.T) {
+	mockLLM := &mockTaskExtHandlerLLM{
+		response: &llm.ChatResponse{Choices: []llm.ChatChoice{{Message: llm.ChatMessage{Role: "assistant", Content: "billing, urgent"}}}},
+	}
+	handler, db := setupTaskExtendedHandlerTest(t, mockLLM)
+	displayID := uint(77)
+	inbox := &model.Inbox{AccountID: 1, Name: "Inbox", ChannelType: "web_widget"}
+	require.NoError(t, db.Create(inbox).Error)
+	conv := &model.Conversation{AccountID: 1, InboxID: inbox.ID, DisplayID: &displayID, Status: "open", ChannelType: "web_widget", Channel: "web_widget"}
+	require.NoError(t, db.Create(conv).Error)
+	msg := &model.Message{ConversationID: conv.ID, AccountID: 1, InboxID: inbox.ID, SenderType: "contact", Content: "billing help", ContentType: "text", MessageType: "incoming"}
+	require.NoError(t, db.Create(msg).Error)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "1"}}
+	c.Request = httptest.NewRequest(http.MethodPost, "/api/v1/accounts/1/captain/tasks/label_suggestion", bytes.NewReader([]byte(`{"conversation_display_id":77}`)))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.LabelSuggestion(c)
+
+	require.Equal(t, http.StatusOK, w.Code)
+	var resp map[string]interface{}
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
+	assert.Equal(t, "billing, urgent", resp["message"])
+	assert.NotContains(t, resp, "success")
+	require.Contains(t, resp, "follow_up_context")
+
+	var stored []model.CopilotSuggestionMessage
+	require.NoError(t, db.Find(&stored).Error)
+	require.Len(t, stored, 1)
+	assert.Equal(t, "billing, urgent", stored[0].Content)
+}
+
 func TestCaptainTaskExtendedHandler_FollowUp(t *testing.T) {
 	mockLLM := &mockTaskExtHandlerLLM{
 		response: &llm.ChatResponse{
@@ -143,3 +180,42 @@ func TestCaptainTaskExtendedHandler_FollowUp_MissingConversationIDs(t *testing.T
 
 	assert.Equal(t, http.StatusBadRequest, w.Code)
 }
+
+func TestCaptainTaskExtendedHandler_FollowUp_ChatwootPostUpdatesContext(t *testing.T) {
+	mockLLM := &mockTaskExtHandlerLLM{
+		response: &llm.ChatResponse{Choices: []llm.ChatChoice{{Message: llm.ChatMessage{Role: "assistant", Content: "Refined answer"}}}},
+	}
+	handler, db := setupTaskExtendedHandlerTest(t, mockLLM)
+	displayID := uint(88)
+	inbox := &model.Inbox{AccountID: 1, Name: "Inbox", ChannelType: "web_widget"}
+	require.NoError(t, db.Create(inbox).Error)
+	conv := &model.Conversation{AccountID: 1, InboxID: inbox.ID, DisplayID: &displayID, Status: "open", ChannelType: "web_widget", Channel: "web_widget"}
+	require.NoError(t, db.Create(conv).Error)
+
+	body := []byte(`{
+    "conversation_display_id": 88,
+    "message": "Make it warmer",
+    "follow_up_context": {
+      "event_name": "professional",
+      "original_context": "Original draft",
+      "last_response": "Previous answer",
+      "conversation_history": []
+    }
+  }`)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "1"}}
+	c.Request = httptest.NewRequest(http.MethodPost, "/api/v1/accounts/1/captain/tasks/follow_up", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.FollowUp(c)
+
+	require.Equal(t, http.StatusOK, w.Code)
+	var resp map[string]interface{}
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
+	assert.Equal(t, "Refined answer", resp["message"])
+	ctx := resp["follow_up_context"].(map[string]interface{})
+	assert.Equal(t, "Refined answer", ctx["last_response"])
+	history := ctx["conversation_history"].([]interface{})
+	require.Len(t, history, 2)
+}
diff --git a/internal/handler/api/v1/captain_task_handler.go b/internal/handler/api/v1/captain_task_handler.go
index 1d9f961c..8db8e65e 100644
--- a/internal/handler/api/v1/captain_task_handler.go
+++ b/internal/handler/api/v1/captain_task_handler.go
@@ -45,11 +45,11 @@ func (h *CaptainTaskHandler) ReplySuggestion(c *gin.Context) {
 	result, err := h.svc.ReplySuggestion(c.Request.Context(), uint(accountID), &req)
 	if err != nil {
 		applogger.L().Errorf("ReplySuggestion: %v", err)
-		response.AbortWithStatusError(c, http.StatusInternalServerError, response.ErrInternal, "failed to generate reply suggestions")
+		renderCaptainTaskError(c, err)
 		return
 	}
 
-	response.OK(c, result)
+	renderCaptainTaskPayload(c, result.Message, result.FollowUpContext)
 }
 
 // Summarize generates a concise summary of a conversation.
@@ -70,11 +70,11 @@ func (h *CaptainTaskHandler) Summarize(c *gin.Context) {
 	result, err := h.svc.Summarize(c.Request.Context(), uint(accountID), &req)
 	if err != nil {
 		applogger.L().Errorf("Summarize: %v", err)
-		response.AbortWithStatusError(c, http.StatusInternalServerError, response.ErrInternal, "failed to summarize conversation")
+		renderCaptainTaskError(c, err)
 		return
 	}
 
-	response.OK(c, result)
+	renderCaptainTaskPayload(c, result.Message, result.FollowUpContext)
 }
 
 // Rewrite rewrites a draft message to improve tone, clarity, or language.
@@ -95,11 +95,11 @@ func (h *CaptainTaskHandler) Rewrite(c *gin.Context) {
 	result, err := h.svc.Rewrite(c.Request.Context(), uint(accountID), &req)
 	if err != nil {
 		applogger.L().Errorf("Rewrite: %v", err)
-		response.AbortWithStatusError(c, http.StatusInternalServerError, response.ErrInternal, "failed to rewrite message")
+		renderCaptainTaskError(c, err)
 		return
 	}
 
-	response.OK(c, result)
+	renderCaptainTaskPayload(c, result.Message, result.FollowUpContext)
 }
 
 // --- SSE Streaming Endpoints (M12) ---
@@ -280,3 +280,20 @@ func captainEscapeJSONString(s string) string {
 	}
 	return result.String()
 }
+
+func renderCaptainTaskPayload(c *gin.Context, message string, followUpContext map[string]interface{}) {
+	payload := gin.H{"message": message}
+	if followUpContext != nil {
+		payload["follow_up_context"] = followUpContext
+	}
+	c.JSON(http.StatusOK, payload)
+}
+
+func renderCaptainTaskError(c *gin.Context, err error) {
+	status, message, ok := service.CaptainTaskErrorStatus(err)
+	if !ok {
+		status = http.StatusUnprocessableEntity
+		message = err.Error()
+	}
+	c.JSON(status, gin.H{"error": message})
+}
diff --git a/internal/handler/api/v1/captain_task_handler_test.go b/internal/handler/api/v1/captain_task_handler_test.go
new file mode 100644
index 00000000..b6c6a4ca
--- /dev/null
+++ b/internal/handler/api/v1/captain_task_handler_test.go
@@ -0,0 +1,103 @@
+package v1
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/gochat/gochat/internal/llm"
+	"github.com/gochat/gochat/internal/model"
+	"github.com/gochat/gochat/internal/repository"
+	"github.com/gochat/gochat/internal/service"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/driver/sqlite"
+	"gorm.io/gorm"
+)
+
+type mockCaptainTaskHandlerLLM struct {
+	response *llm.ChatResponse
+	err      error
+}
+
+func (m *mockCaptainTaskHandlerLLM) ChatCompletion(_ context.Context, _ llm.ChatRequest) (*llm.ChatResponse, error) {
+	return m.response, m.err
+}
+func (m *mockCaptainTaskHandlerLLM) CreateEmbedding(_ context.Context, _ llm.EmbeddingRequest) (*llm.EmbeddingResponse, error) {
+	return nil, nil
+}
+func (m *mockCaptainTaskHandlerLLM) ChatCompletionStream(_ context.Context, _ llm.ChatRequest, _ func(llm.StreamChunk) error) error {
+	return nil
+}
+
+func setupCaptainTaskHandlerTest(t *testing.T, provider llm.Provider) (*CaptainTaskHandler, *gorm.DB) {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	dbName := fmt.Sprintf("file:%s?mode=memory&cache=private", t.Name())
+	db, err := gorm.Open(sqlite.Open(dbName), &gorm.Config{})
+	require.NoError(t, err)
+	require.NoError(t, db.AutoMigrate(
+		&model.Account{},
+		&model.Conversation{},
+		&model.Message{},
+		&model.CaptainAssistant{},
+		&model.CaptainAssistantResponse{},
+		&model.CaptainCustomTool{},
+		&model.CaptainDocument{},
+		&model.CopilotSuggestionMessage{},
+	))
+	assistantRepo := repository.NewCaptainAssistantRepo(db)
+	responseRepo := repository.NewCaptainAssistantResponseRepo(db)
+	customToolRepo := repository.NewCaptainCustomToolRepo(db)
+	conversationRepo := repository.NewConversationRepo(db)
+	messageRepo := repository.NewMessageRepo(db)
+	suggestionRepo := repository.NewCopilotSuggestionRepo(db)
+	svc := service.NewCaptainTaskService(assistantRepo, responseRepo, customToolRepo, conversationRepo, messageRepo, provider, nil, suggestionRepo)
+	return NewCaptainTaskHandler(svc), db
+}
+
+func TestCaptainTaskHandler_Summarize_ChatwootRawPayload(t *testing.T) {
+	provider := &mockCaptainTaskHandlerLLM{response: &llm.ChatResponse{Choices: []llm.ChatChoice{{Message: llm.ChatMessage{Role: "assistant", Content: "Short summary"}}}}}
+	handler, db := setupCaptainTaskHandlerTest(t, provider)
+	displayID := uint(123)
+	conv := &model.Conversation{AccountID: 1, DisplayID: &displayID, Status: "open", ChannelType: "web_widget", Channel: "web_widget"}
+	require.NoError(t, db.Create(conv).Error)
+	require.NoError(t, db.Create(&model.Message{ConversationID: conv.ID, AccountID: 1, SenderType: "contact", MessageType: "incoming", Content: "Need help"}).Error)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "1"}}
+	c.Request = httptest.NewRequest(http.MethodPost, "/api/v1/accounts/1/captain/tasks/summarize", bytes.NewReader([]byte(`{"conversation_display_id":123}`)))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Summarize(c)
+
+	require.Equal(t, http.StatusOK, w.Code)
+	var resp map[string]interface{}
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
+	assert.Equal(t, "Short summary", resp["message"])
+	assert.NotContains(t, resp, "success")
+}
+
+func TestCaptainTaskHandler_Rewrite_NoProviderRawDisabled(t *testing.T) {
+	handler, _ := setupCaptainTaskHandlerTest(t, nil)
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	c.Params = gin.Params{{Key: "id", Value: "1"}}
+	c.Request = httptest.NewRequest(http.MethodPost, "/api/v1/accounts/1/captain/tasks/rewrite", bytes.NewReader([]byte(`{"content":"hello","operation":"professional"}`)))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	handler.Rewrite(c)
+
+	require.Equal(t, http.StatusUnprocessableEntity, w.Code)
+	var resp map[string]interface{}
+	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &resp))
+	assert.Equal(t, "Captain is disabled", resp["error"])
+	assert.NotContains(t, resp, "success")
+}
diff --git a/internal/model/copilot_models.go b/internal/model/copilot_models.go
index 21e32e05..3ea8dfc6 100644
--- a/internal/model/copilot_models.go
+++ b/internal/model/copilot_models.go
@@ -2,6 +2,9 @@ package model
 
 import (
 	"encoding/json"
+	"fmt"
+
+	"gorm.io/gorm"
 )
 
 // --- Copilot Thread Model ---
@@ -62,6 +65,30 @@ type CopilotMessage struct {
 
 func (CopilotMessage) TableName() string { return "copilot_messages" }
 
+// BeforeSave mirrors Chatwoot's CopilotMessage JSON key validation so
+// reloadable tool-call state stays compatible with the reused frontend.
+func (m *CopilotMessage) BeforeSave(tx *gorm.DB) error {
+	if len(m.Message) == 0 {
+		return nil
+	}
+	var msgMap map[string]interface{}
+	if err := json.Unmarshal(m.Message, &msgMap); err != nil {
+		return err
+	}
+	allowed := map[string]bool{
+		"content":          true,
+		"reasoning":        true,
+		"function_name":    true,
+		"reply_suggestion": true,
+	}
+	for key := range msgMap {
+		if !allowed[key] {
+			return fmt.Errorf("message contains invalid attribute: %s", key)
+		}
+	}
+	return nil
+}
+
 // GetMessageContent extracts the "content" field from the JSONB message.
 func (m *CopilotMessage) GetMessageContent() string {
 	var msgMap map[string]interface{}
diff --git a/internal/repository/copilot_message_repo_test.go b/internal/repository/copilot_message_repo_test.go
index ed950c55..e425ea0c 100644
--- a/internal/repository/copilot_message_repo_test.go
+++ b/internal/repository/copilot_message_repo_test.go
@@ -53,6 +53,47 @@ func TestCopilotMessageRepo_Create(t *testing.T) {
 	assert.Equal(t, model.CopilotMessageTypeUser, msg.MessageType)
 }
 
+func TestCopilotMessageRepo_Create_AllowsChatwootMessageKeys(t *testing.T) {
+	db := setupTestDB(t, &model.CopilotThread{}, &model.CopilotMessage{})
+	repo := NewCopilotMessageRepo(db)
+	thread := createTestCopilotThread(t, db, 1, 1, "Tool Thread")
+	message, _ := json.Marshal(map[string]interface{}{
+		"content":          "Looking up docs",
+		"reasoning":        "Need product data",
+		"function_name":    "search_documentation",
+		"reply_suggestion": "Try this reply",
+	})
+
+	err := repo.Create(context.Background(), &model.CopilotMessage{
+		AccountID:       1,
+		CopilotThreadID: thread.ID,
+		MessageType:     model.CopilotMessageTypeAssistant,
+		Message:         message,
+	})
+
+	require.NoError(t, err)
+}
+
+func TestCopilotMessageRepo_Create_RejectsUnknownMessageKeys(t *testing.T) {
+	db := setupTestDB(t, &model.CopilotThread{}, &model.CopilotMessage{})
+	repo := NewCopilotMessageRepo(db)
+	thread := createTestCopilotThread(t, db, 1, 1, "Bad Tool Thread")
+	message, _ := json.Marshal(map[string]interface{}{
+		"content": "hello",
+		"unknown": "bad",
+	})
+
+	err := repo.Create(context.Background(), &model.CopilotMessage{
+		AccountID:       1,
+		CopilotThreadID: thread.ID,
+		MessageType:     model.CopilotMessageTypeAssistant,
+		Message:         message,
+	})
+
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "invalid attribute: unknown")
+}
+
 // --- 2. GetByID ---
 func TestCopilotMessageRepo_GetByID(t *testing.T) {
 	db := setupTestDB(t, &model.CopilotThread{}, &model.CopilotMessage{})
@@ -181,4 +222,4 @@ func TestCopilotMessageRepo_DeleteByThread(t *testing.T) {
 	require.NoError(t, errAfter)
 	assert.Equal(t, int64(0), countAfter)
 	assert.Len(t, msgsAfter, 0)
-}
\ No newline at end of file
+}
diff --git a/internal/service/captain_task_extended_service.go b/internal/service/captain_task_extended_service.go
index ea4ab23d..0df5a5f2 100644
--- a/internal/service/captain_task_extended_service.go
+++ b/internal/service/captain_task_extended_service.go
@@ -7,9 +7,10 @@ import (
 	"strings"
 
 	"github.com/gochat/gochat/internal/llm"
+	"github.com/gochat/gochat/internal/model"
 	"github.com/gochat/gochat/internal/repository"
-	pkgvalidator "github.com/gochat/gochat/pkg/validator"
 	applogger "github.com/gochat/gochat/pkg/logger"
+	pkgvalidator "github.com/gochat/gochat/pkg/validator"
 )
 
 // CaptainTaskExtendedService provides label suggestion and follow-up features.
@@ -20,6 +21,7 @@ type CaptainTaskExtendedService struct {
 	messageRepo      *repository.MessageRepo
 	assistantRepo    *repository.CaptainAssistantRepo
 	preferenceRepo   *repository.CaptainPreferenceRepo
+	suggestionRepo   *repository.CopilotSuggestionRepo
 	llmProvider      llm.Provider
 }
 
@@ -29,12 +31,18 @@ func NewCaptainTaskExtendedService(
 	assistantRepo *repository.CaptainAssistantRepo,
 	preferenceRepo *repository.CaptainPreferenceRepo,
 	llmProvider llm.Provider,
+	suggestionRepos ...*repository.CopilotSuggestionRepo,
 ) *CaptainTaskExtendedService {
+	var suggestionRepo *repository.CopilotSuggestionRepo
+	if len(suggestionRepos) > 0 {
+		suggestionRepo = suggestionRepos[0]
+	}
 	return &CaptainTaskExtendedService{
 		conversationRepo: conversationRepo,
 		messageRepo:      messageRepo,
 		assistantRepo:    assistantRepo,
 		preferenceRepo:   preferenceRepo,
+		suggestionRepo:   suggestionRepo,
 		llmProvider:      llmProvider,
 	}
 }
@@ -81,6 +89,21 @@ type FollowUpResult struct {
 	FollowUps []FollowUpTask `json:"follow_ups"`
 }
 
+type ChatwootLabelSuggestionRequest struct {
+	ConversationDisplayID uint `json:"conversation_display_id"`
+}
+
+type ChatwootFollowUpRequest struct {
+	FollowUpContext       map[string]interface{} `json:"follow_up_context"`
+	Message               string                 `json:"message"`
+	ConversationDisplayID uint                   `json:"conversation_display_id"`
+}
+
+type ChatwootTaskResult struct {
+	Message         *string                `json:"message"`
+	FollowUpContext map[string]interface{} `json:"follow_up_context,omitempty"`
+}
+
 // --- Helper: fetch conversation context for LLM ---
 
 func (s *CaptainTaskExtendedService) fetchConversationMessages(ctx context.Context, conversationID uint) (string, error) {
@@ -126,6 +149,85 @@ func (s *CaptainTaskExtendedService) buildAssistantContext(ctx context.Context,
 	return ""
 }
 
+func (s *CaptainTaskExtendedService) LabelSuggestion(ctx context.Context, accountID uint, req *ChatwootLabelSuggestionRequest) (*ChatwootTaskResult, error) {
+	if s.llmProvider == nil {
+		return nil, taskError(422, "Captain is disabled")
+	}
+	conversation, err := s.conversationRepo.FindByAccountAndDisplayIDOrID(ctx, accountID, req.ConversationDisplayID)
+	if err != nil {
+		return nil, taskError(422, "Conversation not found")
+	}
+	msgs, err := s.fetchConversationMessages(ctx, conversation.ID)
+	if err != nil {
+		return nil, err
+	}
+	if strings.TrimSpace(msgs) == "" {
+		return &ChatwootTaskResult{Message: nil}, nil
+	}
+
+	llmResp, err := s.llmProvider.ChatCompletion(ctx, llm.ChatRequest{
+		Messages: []llm.ChatMessage{
+			{Role: "system", Content: `Suggest concise labels for this customer support conversation. Return only a comma-separated label list.`},
+			{Role: "user", Content: msgs},
+		},
+		Temperature: 0.2,
+		MaxTokens:   256,
+	})
+	if err != nil {
+		return nil, err
+	}
+	if llmResp == nil || len(llmResp.Choices) == 0 {
+		return &ChatwootTaskResult{Message: nil}, nil
+	}
+	message := strings.TrimSpace(llmResp.Choices[0].Message.Content)
+	if message == "" {
+		return &ChatwootTaskResult{Message: nil}, nil
+	}
+	s.persistChatwootTaskSuggestion(ctx, accountID, conversation.ID, message, model.CopilotSuggestionTypeSuggestion)
+	return &ChatwootTaskResult{
+		Message:         &message,
+		FollowUpContext: buildTaskFollowUpContext("label_suggestion", msgs, message, conversation.ChannelType),
+	}, nil
+}
+
+func (s *CaptainTaskExtendedService) FollowUp(ctx context.Context, accountID uint, req *ChatwootFollowUpRequest) (*ChatwootTaskResult, error) {
+	if s.llmProvider == nil {
+		return nil, taskError(422, "Captain is disabled")
+	}
+	if !validTaskFollowUpContext(req.FollowUpContext) {
+		return nil, taskError(422, "Follow-up context missing")
+	}
+	conversation, err := s.conversationRepo.FindByAccountAndDisplayIDOrID(ctx, accountID, req.ConversationDisplayID)
+	if err != nil {
+		return nil, taskError(422, "Conversation not found")
+	}
+	messages := []llm.ChatMessage{
+		{Role: "system", Content: "You just performed a customer support assistant task. Refine the previous result based on the agent's feedback. Output only the refined reply."},
+		{Role: "user", Content: fmt.Sprint(req.FollowUpContext["original_context"])},
+		{Role: "assistant", Content: fmt.Sprint(req.FollowUpContext["last_response"])},
+	}
+	for _, historyMessage := range followUpHistory(req.FollowUpContext) {
+		messages = append(messages, historyMessage)
+	}
+	messages = append(messages, llm.ChatMessage{Role: "user", Content: req.Message})
+
+	llmResp, err := s.llmProvider.ChatCompletion(ctx, llm.ChatRequest{
+		Messages:    messages,
+		Temperature: 0.3,
+		MaxTokens:   512,
+	})
+	if err != nil {
+		return nil, err
+	}
+	if llmResp == nil || len(llmResp.Choices) == 0 {
+		return nil, taskError(422, "No response from LLM")
+	}
+	message := llmResp.Choices[0].Message.Content
+	updatedContext := updateTaskFollowUpContext(req.FollowUpContext, req.Message, message, conversation.ChannelType)
+	s.persistChatwootTaskSuggestion(ctx, accountID, conversation.ID, message, model.CopilotSuggestionTypeSuggestion)
+	return &ChatwootTaskResult{Message: &message, FollowUpContext: updatedContext}, nil
+}
+
 // --- Label Suggestion (GET) ---
 // Reference: Chatwoot Captain::ConversationInsightController#suggest_labels
 
@@ -174,7 +276,7 @@ Respond in JSON format:
 }` + assistantCtx + promptSuffix
 
 		llmResp, err := s.llmProvider.ChatCompletion(ctx, llm.ChatRequest{
-			Model:       "",
+			Model: "",
 			Messages: []llm.ChatMessage{
 				{Role: "system", Content: systemPrompt},
 				{Role: "user", Content: msgs},
@@ -278,7 +380,7 @@ Respond in JSON format:
 }` + statusHint + assistantCtx + promptSuffix
 
 		llmResp, err := s.llmProvider.ChatCompletion(ctx, llm.ChatRequest{
-			Model:       "",
+			Model: "",
 			Messages: []llm.ChatMessage{
 				{Role: "system", Content: systemPrompt},
 				{Role: "user", Content: msgs},
@@ -321,3 +423,102 @@ Respond in JSON format:
 
 	return &FollowUpResult{FollowUps: followUps}, nil
 }
+
+func buildTaskFollowUpContext(eventName, originalContext, lastResponse, channelType string) map[string]interface{} {
+	ctx := map[string]interface{}{
+		"event_name":           eventName,
+		"original_context":     originalContext,
+		"last_response":        lastResponse,
+		"conversation_history": []map[string]string{},
+	}
+	if channelType != "" {
+		ctx["channel_type"] = channelType
+	}
+	return ctx
+}
+
+func validTaskFollowUpContext(ctx map[string]interface{}) bool {
+	if ctx == nil {
+		return false
+	}
+	eventName, _ := ctx["event_name"].(string)
+	if !allowedTaskFollowUpEvent(eventName) {
+		return false
+	}
+	return strings.TrimSpace(fmt.Sprint(ctx["original_context"])) != "" && strings.TrimSpace(fmt.Sprint(ctx["last_response"])) != ""
+}
+
+func allowedTaskFollowUpEvent(eventName string) bool {
+	switch eventName {
+	case "professional", "casual", "friendly", "confident", "straightforward", "fix_spelling_grammar", "improve", "summarize", "reply_suggestion", "label_suggestion":
+		return true
+	default:
+		return false
+	}
+}
+
+func followUpHistory(ctx map[string]interface{}) []llm.ChatMessage {
+	raw, ok := ctx["conversation_history"].([]interface{})
+	if !ok {
+		return nil
+	}
+	messages := make([]llm.ChatMessage, 0, len(raw))
+	for _, item := range raw {
+		m, ok := item.(map[string]interface{})
+		if !ok {
+			continue
+		}
+		role := fmt.Sprint(m["role"])
+		content := fmt.Sprint(m["content"])
+		if role == "" || content == "" {
+			continue
+		}
+		messages = append(messages, llm.ChatMessage{Role: role, Content: content})
+	}
+	return messages
+}
+
+func updateTaskFollowUpContext(ctx map[string]interface{}, userMessage, assistantMessage, channelType string) map[string]interface{} {
+	history := make([]map[string]string, 0)
+	if raw, ok := ctx["conversation_history"].([]interface{}); ok {
+		for _, item := range raw {
+			m, ok := item.(map[string]interface{})
+			if !ok {
+				continue
+			}
+			history = append(history, map[string]string{"role": fmt.Sprint(m["role"]), "content": fmt.Sprint(m["content"])})
+		}
+	}
+	history = append(history,
+		map[string]string{"role": "user", "content": userMessage},
+		map[string]string{"role": "assistant", "content": assistantMessage},
+	)
+	updated := map[string]interface{}{
+		"event_name":           ctx["event_name"],
+		"original_context":     ctx["original_context"],
+		"last_response":        assistantMessage,
+		"conversation_history": history,
+	}
+	if channelType != "" {
+		updated["channel_type"] = channelType
+	} else if value, ok := ctx["channel_type"]; ok {
+		updated["channel_type"] = value
+	}
+	return updated
+}
+
+func (s *CaptainTaskExtendedService) persistChatwootTaskSuggestion(ctx context.Context, accountID, conversationID uint, content string, suggestionType model.CopilotSuggestionType) {
+	if s.suggestionRepo == nil || strings.TrimSpace(content) == "" {
+		return
+	}
+	msg := &model.CopilotSuggestionMessage{
+		AccountID:      accountID,
+		ConversationID: conversationID,
+		Content:        content,
+		SuggestionType: suggestionType,
+		Status:         model.CopilotSuggestionStatusPending,
+	}
+	if err := s.suggestionRepo.Create(ctx, msg); err != nil {
+		applogger.L().Warnf("Captain extended task suggestion persistence failed: %v", err)
+	}
+}
diff --git a/internal/service/captain_task_service.go b/internal/service/captain_task_service.go
index ad58af40..bdc404bd 100644
--- a/internal/service/captain_task_service.go
+++ b/internal/service/captain_task_service.go
@@ -8,6 +8,7 @@ import (
 	"github.com/pgvector/pgvector-go"
 
 	"github.com/gochat/gochat/internal/llm"
+	"github.com/gochat/gochat/internal/model"
 	"github.com/gochat/gochat/internal/repository"
 	applogger "github.com/gochat/gochat/pkg/logger"
 )
@@ -17,14 +18,15 @@ import (
 // These tasks (reply_suggestion, summarize, rewrite) are independent of
 // CopilotThread flow — they provide one-shot AI suggestions for agents.
 type CaptainTaskService struct {
-	assistantRepo          *repository.CaptainAssistantRepo
-	responseRepo           *repository.CaptainAssistantResponseRepo
-	customToolRepo         *repository.CaptainCustomToolRepo
-	conversationRepo       *repository.ConversationRepo
-	messageRepo            *repository.MessageRepo
-	llmProvider            llm.Provider
-	promptBuilder          *SystemPromptBuilder
-	copilotContextSvc      *CopilotContextService // M12: context enrichment for richer prompts
+	assistantRepo     *repository.CaptainAssistantRepo
+	responseRepo      *repository.CaptainAssistantResponseRepo
+	customToolRepo    *repository.CaptainCustomToolRepo
+	conversationRepo  *repository.ConversationRepo
+	messageRepo       *repository.MessageRepo
+	suggestionRepo    *repository.CopilotSuggestionRepo
+	llmProvider       llm.Provider
+	promptBuilder     *SystemPromptBuilder
+	copilotContextSvc *CopilotContextService // M12: context enrichment for richer prompts
 }
 
 // NewCaptainTaskService creates a new CaptainTaskService.
@@ -36,16 +38,22 @@ func NewCaptainTaskService(
 	messageRepo *repository.MessageRepo,
 	llmProvider llm.Provider,
 	copilotContextSvc *CopilotContextService,
+	suggestionRepos ...*repository.CopilotSuggestionRepo,
 ) *CaptainTaskService {
+	var suggestionRepo *repository.CopilotSuggestionRepo
+	if len(suggestionRepos) > 0 {
+		suggestionRepo = suggestionRepos[0]
+	}
 	return &CaptainTaskService{
-		assistantRepo:      assistantRepo,
-		responseRepo:       responseRepo,
-		customToolRepo:     customToolRepo,
-		conversationRepo:   conversationRepo,
-		messageRepo:        messageRepo,
-		llmProvider:        llmProvider,
-		promptBuilder:      NewSystemPromptBuilder(),
-		copilotContextSvc:  copilotContextSvc,
+		assistantRepo:     assistantRepo,
+		responseRepo:      responseRepo,
+		customToolRepo:    customToolRepo,
+		conversationRepo:  conversationRepo,
+		messageRepo:       messageRepo,
+		suggestionRepo:    suggestionRepo,
+		llmProvider:       llmProvider,
+		promptBuilder:     NewSystemPromptBuilder(),
+		copilotContextSvc: copilotContextSvc,
 	}
 }
 
@@ -53,40 +61,71 @@ func NewCaptainTaskService(
 
 // TaskReplySuggestionRequest is the DTO for reply_suggestion task.
 type TaskReplySuggestionRequest struct {
-	AssistantID    uint   `json:"assistant_id" validate:"required"`
-	ConversationID uint   `json:"conversation_id" validate:"required"`
-	Tone           string `json:"tone"` // e.g. "friendly", "formal", "professional"
+	AssistantID           uint   `json:"assistant_id"`
+	ConversationID        uint   `json:"conversation_id"`
+	ConversationDisplayID uint   `json:"conversation_display_id"`
+	Tone                  string `json:"tone"` // legacy local compatibility
 }
 
 // TaskSummarizeRequest is the DTO for summarize task.
 type TaskSummarizeRequest struct {
-	ConversationID uint   `json:"conversation_id" validate:"required"`
-	Language       string `json:"language"` // summary output language, default "zh-CN"
+	ConversationID        uint   `json:"conversation_id"`
+	ConversationDisplayID uint   `json:"conversation_display_id"`
+	Language              string `json:"language"` // legacy local compatibility
 }
 
 // TaskRewriteRequest is the DTO for rewrite task.
 type TaskRewriteRequest struct {
-	Content     string `json:"content" validate:"required"`
-	Tone        string `json:"tone"`        // desired tone, e.g. "friendly", "formal"
-	Language    string `json:"language"`    // desired output language
-	AssistantID uint   `json:"assistant_id"` // optional: use assistant style guidelines
+	Content               string `json:"content" validate:"required"`
+	Operation             string `json:"operation"`
+	ConversationDisplayID uint   `json:"conversation_display_id"`
+	Tone                  string `json:"tone"`         // legacy local compatibility
+	Language              string `json:"language"`     // legacy local compatibility
+	AssistantID           uint   `json:"assistant_id"` // legacy local compatibility
 }
 
 // TaskReplySuggestionResult holds the reply suggestion result.
 type TaskReplySuggestionResult struct {
-	Suggestions []string `json:"suggestions"`
-	Source      string   `json:"source"` // "llm", "rag+llm"
+	Message         string                 `json:"message"`
+	FollowUpContext map[string]interface{} `json:"follow_up_context,omitempty"`
+	Suggestions     []string               `json:"suggestions"`
+	Source          string                 `json:"source"` // "llm", "rag+llm"
 }
 
 // TaskSummarizeResult holds the conversation summary result.
 type TaskSummarizeResult struct {
-	Summary string `json:"summary"`
+	Message         string                 `json:"message"`
+	FollowUpContext map[string]interface{} `json:"follow_up_context,omitempty"`
+	Summary         string                 `json:"summary"`
 }
 
 // TaskRewriteResult holds the rewritten content result.
 type TaskRewriteResult struct {
-	RewrittenContent string `json:"rewritten_content"`
-	Tone             string `json:"tone"`
+	Message          string                 `json:"message"`
+	FollowUpContext  map[string]interface{} `json:"follow_up_context,omitempty"`
+	RewrittenContent string                 `json:"rewritten_content"`
+	Tone             string                 `json:"tone"`
+}
+
+type CaptainTaskError struct {
+	Status  int
+	Message string
+}
+
+func (e *CaptainTaskError) Error() string { return e.Message }
+
+func taskError(status int, message string) error {
+	return &CaptainTaskError{Status: status, Message: message}
+}
+
+func CaptainTaskErrorStatus(err error) (int, string, bool) {
+	if err == nil {
+		return 0, "", false
+	}
+	if taskErr, ok := err.(*CaptainTaskError); ok {
+		return taskErr.Status, taskErr.Message, true
+	}
+	return 0, "", false
 }
 
 // --- Reply Suggestion Task ---
@@ -100,14 +139,28 @@ type TaskRewriteResult struct {
 // 4. Build prompt with conversation context + relevant FAQ answers
 // 5. Call LLM to generate reply suggestions
 func (s *CaptainTaskService) ReplySuggestion(ctx context.Context, accountID uint, req *TaskReplySuggestionRequest) (*TaskReplySuggestionResult, error) {
-	// Fetch assistant config for prompt building
-	assistant, err := s.assistantRepo.GetByID(ctx, req.AssistantID)
-	if err != nil {
-		return nil, fmt.Errorf("assistant not found: %w", err)
+	if s.llmProvider == nil {
+		return nil, taskError(422, "Captain is disabled")
 	}
 
-	// Fetch conversation messages as context
-	messages, err := s.fetchConversationContext(ctx, req.ConversationID)
+	var assistantCfg *model.CaptainAssistant
+	var err error
+	if req.AssistantID > 0 {
+		assistantCfg, err = s.assistantRepo.GetByAccountAndID(ctx, accountID, req.AssistantID)
+		if err != nil {
+			return nil, taskError(422, "assistant not found")
+		}
+	} else {
+		assistantCfg = &model.CaptainAssistant{AccountID: accountID, Name: "Captain"}
+		assistantCfg.Config = []byte(`{}`)
+	}
+
+	conversation, err := s.resolveTaskConversation(ctx, accountID, req.ConversationDisplayID, req.ConversationID)
+	if err != nil {
+		return nil, taskError(422, "Conversation not found")
+	}
+
+	messages, err := s.fetchConversationContext(ctx, conversation.ID)
 	if err != nil {
 		applogger.L().Warnf("ReplySuggestion: conversation context fetch failed: %v, proceeding without context", err)
 		messages = ""
@@ -115,17 +168,19 @@ func (s *CaptainTaskService) ReplySuggestion(ctx context.Context, accountID uint
 
 	// Search relevant documents via embedding similarity (RAG)
 	ragContext, source := "", "llm"
-	docContext, err := s.searchDocumentation(ctx, req.AssistantID, messages)
-	if err != nil {
-		applogger.L().Warnf("ReplySuggestion: document search failed: %v, proceeding without RAG", err)
-	} else if docContext != "" {
-		ragContext = docContext
-		source = "rag+llm"
+	if req.AssistantID > 0 {
+		docContext, err := s.searchDocumentation(ctx, req.AssistantID, messages)
+		if err != nil {
+			applogger.L().Warnf("ReplySuggestion: document search failed: %v, proceeding without RAG", err)
+		} else if docContext != "" {
+			ragContext = docContext
+			source = "rag+llm"
+		}
 	}
 
 	// Build system prompt using prompt builder
-	cfg, _ := assistant.GetConfig()
-	systemPrompt := s.promptBuilder.BuildReplySuggestionPrompt(assistant, cfg, req.Tone, ragContext)
+	cfg, _ := assistantCfg.GetConfig()
+	systemPrompt := s.promptBuilder.BuildReplySuggestionPrompt(assistantCfg, cfg, req.Tone, ragContext)
 
 	// Build messages for LLM
 	chatMessages := []llm.ChatMessage{
@@ -152,10 +207,17 @@ func (s *CaptainTaskService) ReplySuggestion(ctx context.Context, accountID uint
 
 	// Parse suggestions from LLM response
 	suggestions := parseSuggestions(resp.Choices[0].Message.Content)
+	message := strings.Join(suggestions, "\n")
+	if message == "" {
+		message = resp.Choices[0].Message.Content
+	}
+	s.persistTaskSuggestion(ctx, accountID, conversation.ID, message, model.CopilotSuggestionTypeReply)
 
 	return &TaskReplySuggestionResult{
-		Suggestions: suggestions,
-		Source:      source,
+		Message:         message,
+		FollowUpContext: s.buildFollowUpContext("reply_suggestion", messages, message, conversation),
+		Suggestions:     suggestions,
+		Source:          source,
 	}, nil
 }
 
@@ -164,14 +226,21 @@ func (s *CaptainTaskService) ReplySuggestion(ctx context.Context, accountID uint
 
 // Summarize generates a concise summary of a conversation.
 func (s *CaptainTaskService) Summarize(ctx context.Context, accountID uint, req *TaskSummarizeRequest) (*TaskSummarizeResult, error) {
+	if s.llmProvider == nil {
+		return nil, taskError(422, "Captain is disabled")
+	}
+	conversation, err := s.resolveTaskConversation(ctx, accountID, req.ConversationDisplayID, req.ConversationID)
+	if err != nil {
+		return nil, taskError(422, "Conversation not found")
+	}
 	// Fetch conversation messages
-	messages, err := s.fetchConversationContext(ctx, req.ConversationID)
+	messages, err := s.fetchConversationContext(ctx, conversation.ID)
 	if err != nil {
 		return nil, fmt.Errorf("fetch conversation context: %w", err)
 	}
 
 	if messages == "" {
-		return nil, fmt.Errorf("no messages found for conversation %d", req.ConversationID)
+		return nil, taskError(422, fmt.Sprintf("no messages found for conversation %d", conversation.ID))
 	}
 
 	// Determine output language
@@ -204,8 +273,13 @@ func (s *CaptainTaskService) Summarize(ctx context.Context, accountID uint, req
 		return nil, fmt.Errorf("no response from LLM")
 	}
 
+	message := resp.Choices[0].Message.Content
+	s.persistTaskSuggestion(ctx, accountID, conversation.ID, message, model.CopilotSuggestionTypeSummary)
+
 	return &TaskSummarizeResult{
-		Summary: resp.Choices[0].Message.Content,
+		Message:         message,
+		FollowUpContext: s.buildFollowUpContext("summarize", messages, message, conversation),
+		Summary:         message,
 	}, nil
 }
 
@@ -214,9 +288,18 @@ func (s *CaptainTaskService) Summarize(ctx context.Context, accountID uint, req
 
 // Rewrite rewrites a draft message to improve tone, clarity, or language.
 func (s *CaptainTaskService) Rewrite(ctx context.Context, accountID uint, req *TaskRewriteRequest) (*TaskRewriteResult, error) {
-	tone := req.Tone
-	if tone == "" {
-		tone = "professional"
+	if s.llmProvider == nil {
+		return nil, taskError(422, "Captain is disabled")
+	}
+	operation := req.Operation
+	if operation == "" {
+		operation = req.Tone
+	}
+	if operation == "" {
+		operation = "professional"
+	}
+	if !isAllowedRewriteOperation(operation) {
+		return nil, taskError(422, "Invalid operation: "+operation)
 	}
 
 	language := req.Language
@@ -224,10 +307,22 @@ func (s *CaptainTaskService) Rewrite(ctx context.Context, accountID uint, req *T
 		language = "zh-CN"
 	}
 
-	// Optionally use assistant guidelines for style
 	styleGuidelines := ""
+	var conversation *model.Conversation
+	conversationContext := ""
+	if req.ConversationDisplayID > 0 {
+		resolved, err := s.resolveTaskConversation(ctx, accountID, req.ConversationDisplayID, 0)
+		if err != nil {
+			return nil, taskError(422, "Conversation not found")
+		}
+		conversation = resolved
+		if operation == "improve" {
+			conversationContext, _ = s.fetchConversationContext(ctx, conversation.ID)
+		}
+	}
+
 	if req.AssistantID > 0 {
-		assistant, err := s.assistantRepo.GetByID(ctx, req.AssistantID)
+		assistant, err := s.assistantRepo.GetByAccountAndID(ctx, accountID, req.AssistantID)
 		if err == nil {
 			cfg, _ := assistant.GetConfig()
 			if cfg.Instructions != "" {
@@ -238,9 +333,12 @@ func (s *CaptainTaskService) Rewrite(ctx context.Context, accountID uint, req *T
 			}
 		}
 	}
+	if conversationContext != "" {
+		styleGuidelines += "\nConversation context: " + conversationContext
+	}
 
 	// Build rewrite prompt using prompt builder
-	systemPrompt := s.promptBuilder.BuildRewritePrompt(tone, language, styleGuidelines)
+	systemPrompt := s.promptBuilder.BuildRewritePrompt(operation, language, styleGuidelines)
 
 	chatMessages := []llm.ChatMessage{
 		{Role: "system", Content: systemPrompt},
@@ -263,9 +361,16 @@ func (s *CaptainTaskService) Rewrite(ctx context.Context, accountID uint, req *T
 		return nil, fmt.Errorf("no response from LLM")
 	}
 
+	message := resp.Choices[0].Message.Content
+	if conversation != nil {
+		s.persistTaskSuggestion(ctx, accountID, conversation.ID, message, model.CopilotSuggestionTypeSuggestion)
+	}
+
 	return &TaskRewriteResult{
-		RewrittenContent: resp.Choices[0].Message.Content,
-		Tone:             tone,
+		Message:          message,
+		FollowUpContext:  s.buildFollowUpContext(operation, req.Content, message, conversation),
+		RewrittenContent: message,
+		Tone:             operation,
 	}, nil
 }
 
@@ -294,6 +399,54 @@ func (s *CaptainTaskService) fetchConversationContext(ctx context.Context, conve
 	return builder.String(), nil
 }
 
+func (s *CaptainTaskService) resolveTaskConversation(ctx context.Context, accountID, displayID, legacyID uint) (*model.Conversation, error) {
+	if displayID > 0 {
+		return s.conversationRepo.FindByAccountAndDisplayIDOrID(ctx, accountID, displayID)
+	}
+	if legacyID > 0 {
+		return s.conversationRepo.FindByAccountAndDisplayIDOrID(ctx, accountID, legacyID)
+	}
+	return nil, fmt.Errorf("conversation_display_id required")
+}
+
+func isAllowedRewriteOperation(operation string) bool {
+	switch operation {
+	case "fix_spelling_grammar", "improve", "casual", "professional", "friendly", "confident", "straightforward":
+		return true
+	default:
+		return false
+	}
+}
+
+func (s *CaptainTaskService) buildFollowUpContext(eventName, originalContext, lastResponse string, conversation *model.Conversation) map[string]interface{} {
+	ctx := map[string]interface{}{
+		"event_name":           eventName,
+		"original_context":     originalContext,
+		"last_response":        lastResponse,
+		"conversation_history": []map[string]string{},
+	}
+	if conversation != nil {
+		ctx["channel_type"] = conversation.ChannelType
+	}
+	return ctx
+}
+
+func (s *CaptainTaskService) persistTaskSuggestion(ctx context.Context, accountID, conversationID uint, content string, suggestionType model.CopilotSuggestionType) {
+	if s.suggestionRepo == nil || strings.TrimSpace(content) == "" || conversationID == 0 {
+		return
+	}
+	msg := &model.CopilotSuggestionMessage{
+		AccountID:      accountID,
+		ConversationID: conversationID,
+		Content:        content,
+		SuggestionType: suggestionType,
+		Status:         model.CopilotSuggestionStatusPending,
+	}
+	if err := s.suggestionRepo.Create(ctx, msg); err != nil {
+		applogger.L().Warnf("Captain task suggestion persistence failed: %v", err)
+	}
+}
+
 // searchDocumentation performs embedding-based similarity search on assistant responses.
 // Reference: Chatwoot Captain::Llm::SearchReplyDocumentationService
 func (s *CaptainTaskService) searchDocumentation(ctx context.Context, assistantID uint, query string) (string, error) {
diff --git a/internal/service/captain_task_service_test.go b/internal/service/captain_task_service_test.go
index 749959d0..cc60bec7 100644
--- a/internal/service/captain_task_service_test.go
+++ b/internal/service/captain_task_service_test.go
@@ -77,6 +77,7 @@ func setupCaptainTaskService(t *testing.T) (*gorm.DB, *mockLLMProvider, *Captain
 		&model.CaptainAssistantResponse{},
 		&model.CaptainCustomTool{},
 		&model.CaptainDocument{},
+		&model.CopilotSuggestionMessage{},
 	); err != nil {
 		t.Fatalf("无法自动迁移模型: %v", err)
 	}
@@ -92,12 +93,13 @@ func setupCaptainTaskService(t *testing.T) (*gorm.DB, *mockLLMProvider, *Captain
 	customToolRepo := repository.NewCaptainCustomToolRepo(db)
 	conversationRepo := repository.NewConversationRepo(db)
 	messageRepo := repository.NewMessageRepo(db)
+	suggestionRepo := repository.NewCopilotSuggestionRepo(db)
 	contactRepo := repository.NewContactRepo(db)
 
 	// M12: CopilotContextService for context enrichment in streaming
 	copilotContextSvc := NewCopilotContextService(messageRepo, conversationRepo, contactRepo, mockProvider)
 
-	svc := NewCaptainTaskService(assistantRepo, responseRepo, customToolRepo, conversationRepo, messageRepo, mockProvider, copilotContextSvc)
+	svc := NewCaptainTaskService(assistantRepo, responseRepo, customToolRepo, conversationRepo, messageRepo, mockProvider, copilotContextSvc, suggestionRepo)
 
 	return db, mockProvider, svc
 }
@@ -121,8 +123,10 @@ func createTestAssistant(t *testing.T, db *gorm.DB, accountID uint, name string)
 func createTestConversationWithMessages(t *testing.T, db *gorm.DB, accountID uint, msgs []struct{ sender, content string }) *model.Conversation {
 	t.Helper()
 	conv := &model.Conversation{
-		AccountID: accountID,
-		Status:    "open",
+		AccountID:   accountID,
+		Status:      "open",
+		ChannelType: "web_widget",
+		Channel:     "web_widget",
 	}
 	if err := db.Create(conv).Error; err != nil {
 		t.Fatalf("无法创建测试会话: %v", err)
@@ -199,6 +203,60 @@ func TestCaptainTaskService_Summarize_成功(t *testing.T) {
 	assert.NotEmpty(t, result.Summary)
 }
 
+func TestCaptainTaskService_Summarize_ChatwootDisplayIDPersistsSuggestion(t *testing.T) {
+	db, mockProvider, svc := setupCaptainTaskService(t)
+	account := createTestAccount(t, db)
+	displayID := uint(42)
+	conv := createTestConversationWithMessages(t, db, account.ID, []struct{ sender, content string }{
+		{"contact", "I need help with my invoice"},
+		{"agent", "I can help with billing"},
+	})
+	require.NoError(t, db.Model(conv).Update("display_id", displayID).Error)
+
+	mockProvider.chatResponse = makeChatResponse("Customer needs billing help")
+
+	result, err := svc.Summarize(context.Background(), account.ID, &TaskSummarizeRequest{
+		ConversationDisplayID: displayID,
+	})
+
+	require.NoError(t, err)
+	assert.Equal(t, "Customer needs billing help", result.Message)
+	require.NotNil(t, result.FollowUpContext)
+	assert.Equal(t, "summarize", result.FollowUpContext["event_name"])
+
+	var stored []model.CopilotSuggestionMessage
+	require.NoError(t, db.Where("account_id = ? AND conversation_id = ?", account.ID, conv.ID).Find(&stored).Error)
+	require.Len(t, stored, 1)
+	assert.Equal(t, "Customer needs billing help", stored[0].Content)
+	assert.Equal(t, model.CopilotSuggestionTypeSummary, stored[0].SuggestionType)
+}
+
+func TestCaptainTaskService_Rewrite_InvalidOperation(t *testing.T) {
+	_, _, svc := setupCaptainTaskService(t)
+
+	result, err := svc.Rewrite(context.Background(), 1, &TaskRewriteRequest{Content: "hello", Operation: "pirate"})
+
+	assert.Nil(t, result)
+	status, message, ok := CaptainTaskErrorStatus(err)
+	require.True(t, ok)
+	assert.Equal(t, 422, status)
+	assert.Equal(t, "Invalid operation: pirate", message)
+}
+
+func TestCaptainTaskService_Rewrite_NoProviderDisabled(t *testing.T) {
+	db, _, svc := setupCaptainTaskService(t)
+	svc.llmProvider = nil
+	account := createTestAccount(t, db)
+
+	result, err := svc.Rewrite(context.Background(), account.ID, &TaskRewriteRequest{Content: "hello", Operation: "professional"})
+
+	assert.Nil(t, result)
+	status, message, ok := CaptainTaskErrorStatus(err)
+	require.True(t, ok)
+	assert.Equal(t, 422, status)
+	assert.Equal(t, "Captain is disabled", message)
+}
+
 func TestCaptainTaskService_Summarize_无消息(t *testing.T) {
 	db, _, svc := setupCaptainTaskService(t)
 	account := createTestAccount(t, db)
@@ -416,4 +474,4 @@ func TestParseSuggestions_空内容(t *testing.T) {
 	content := ""
 	result := parseSuggestions(content)
 	assert.Len(t, result, 1)
-}
\ No newline at end of file
+}