Merge pull request #52 from BetaGoRobot/feature-agentic

✨ feat(agent-runtime): 新增金融工具动态发现与注入能力

Author: kevinmatthe

.gitignore

@@ -6,10 +6,11 @@ __debug_bin*
 !.codex/skills/
 !.codex/skills/lark-card-debug/
 !.codex/skills/lark-card-debug/**
+.worktrees/
 .cache
 .env
 *.test
 script/neteaseapi/api_repo/*
 **__pycache__**
 .gocache
-.gomodcache
+.gomodcache

cmd/larkrobot/bootstrap.go

@@ -147,9 +147,10 @@ func newAppComponents(cfg *infraConfig.BaseConfig) *appComponents {
 	agentruntime.SetRuntimeAgenticCutoverBuilder(runtimecutover.BuildDefaultHandler)
 	// agentruntime.SetRuntimeStandardCutoverBuilder(runtimecutover.BuildDefaultStandardHandler)
 	agentruntime.SetDefaultChatToolProvider(handlers.BuildLarkTools)
+	agentruntime.SetDefaultChatCapabilityToolProvider(handlers.BuildRuntimeCapabilityTools)
 	agentruntime.SetChatGenerationPlanExecutor(agentruntime.NewDefaultChatGenerationPlanExecutor())
 	runtimewire.SetDefaultCapabilityProvider(func() []agentruntime.Capability {
-		return capdef.BuildToolCapabilities(handlers.BuildLarkTools(), nil, (*larkim.P2MessageReceiveV1)(nil))
+		return capdef.BuildToolCapabilities(handlers.BuildRuntimeCapabilityTools(), nil, (*larkim.P2MessageReceiveV1)(nil))
 	})
 	messageProcessor := messages.NewMessageProcessor(appconfig.GetManager())
 	reactionProcessor := reaction.NewReactionProcessor()

docs/superpowers/plans/2026-03-26-chat-passive-ratelimit-v2.md

@@ -0,0 +1,217 @@
+# Chat Passive Ratelimit V2 Implementation Plan
+
+> **For agentic workers:** REQUIRED: Use superpowers:subagent-driven-development (if subagents available) or superpowers:executing-plans to implement this plan. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Rework passive chat reply policy so direct mentions/private chat always bypass blocking, passive replies and active interjections use willingness-aware rate limiting, and `search_history` supports richer metadata filters while remaining strictly scoped to the current `chat_id`.
+
+**Architecture:** Split reply policy into three layers: intent policy classifies direct/passive/interject decisions and willingness metadata, ratelimit enforces passive/interject budgets with separate hard/soft accounting, and chat generation narrows default context while relying on explicit history tools for broader retrieval. `search_history` becomes the primary metadata-aware history tool, but the handler owns the `chat_id` boundary so tool calls cannot escape the current conversation.
+
+**Tech Stack:** Go, OpenSearch, Redis, GORM/gen query layer, Lark handlers/operators, Ark Responses tool calling, Go test.
+
+---
+
+## Chunk 1: Metadata-Aware History Search
+
+### Task 1: Extend `search_history` request model and tests
+
+**Files:**
+- Create: `internal/application/lark/history/search_test.go`
+- Modify: `internal/application/lark/history/search.go`
+- Modify: `internal/application/lark/handlers/history_search_handler.go`
+- Modify: `internal/application/lark/handlers/tools_test.go`
+
+- [x] **Step 1: Write failing history-search query builder tests**
+
+Add table-driven tests that verify:
+- `HybridSearchRequest.ChatID` is mandatory for tool usage.
+- `user_id`, `user_name`, and `message_type` filters become exact filter clauses.
+- keyword search still builds both keyword and vector clauses.
+- invalid or missing `ChatID` never produces an unscoped query.
+
+- [x] **Step 2: Run test to verify it fails**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/history -run 'TestHybridSearch' -count=1`
+Expected: FAIL because metadata filtering and mandatory chat scoping are not implemented.
+
+- [x] **Step 3: Write failing handler/tool tests**
+
+Add tests for:
+- `search_history` tool schema exposes metadata fields such as `user_name` and `message_type`.
+- handler always injects `metaData.ChatID` into the search request.
+- handler rejects empty `metaData.ChatID`.
+
+- [x] **Step 4: Run test to verify it fails**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/handlers -run 'Test(SearchHistory|LarkTools)' -count=1`
+Expected: FAIL because the tool does not yet expose the new metadata fields or enforce empty-chat rejection.
+
+- [x] **Step 5: Implement minimal history search changes**
+
+Implement:
+- richer `HybridSearchRequest` metadata fields,
+- a helper that builds the OpenSearch bool query with mandatory `chat_id`,
+- handler-side enforcement that tool searches are scoped to the current chat only.
+
+- [x] **Step 6: Run focused tests**
+
+Run:
+- `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/history -run 'TestHybridSearch' -count=1`
+- `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/handlers -run 'Test(SearchHistory|LarkTools)' -count=1`
+Expected: PASS.
+
+## Chunk 2: Willingness-Aware Passive Reply Policy
+
+### Task 2: Expand intent analysis payload and tests
+
+**Files:**
+- Modify: `internal/application/lark/intentmeta/types.go`
+- Modify: `internal/application/lark/intent/recognizer.go`
+- Modify: `internal/application/lark/intent/recognizer_test.go`
+- Modify: `internal/application/lark/messages/ops/intent_recognize_op_test.go`
+
+- [x] **Step 1: Write failing intent payload tests**
+
+Add tests that cover:
+- parsing/sanitizing `reply_mode`, `user_willingness`, `interrupt_risk`, `needs_history`, `needs_web`,
+- conservative defaults for missing/invalid values,
+- prompt text explicitly describing direct/passive/interject distinctions.
+
+- [x] **Step 2: Run test to verify it fails**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/intent ./internal/application/lark/messages/ops -run 'Test(Intent|AnalyzeMessage|IntentRecognize)' -count=1`
+Expected: FAIL because the new fields and prompt guidance do not exist yet.
+
+- [x] **Step 3: Implement minimal intent model changes**
+
+Implement:
+- new enum/type definitions in `intentmeta`,
+- updated recognizer system prompt and JSON parsing,
+- sanitization rules that keep direct/private traffic conservative and passive/interject scores bounded to `0-100`.
+
+- [x] **Step 4: Run focused tests**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/intent ./internal/application/lark/messages/ops -run 'Test(Intent|AnalyzeMessage|IntentRecognize)' -count=1`
+Expected: PASS.
+
+### Task 3: Rework passive reply decider and ratelimit accounting
+
+**Files:**
+- Modify: `internal/application/lark/ratelimit/rate_limiter.go`
+- Modify: `internal/application/lark/ratelimit/integration.go`
+- Modify: `internal/application/lark/ratelimit/rate_limiter_test.go`
+- Modify: `internal/application/lark/messages/ops/chat_op.go`
+- Modify: `internal/application/config/accessor.go`
+
+- [x] **Step 1: Write failing decider/ratelimit tests**
+
+Add tests that verify:
+- direct mention traffic is never blocked by `Allow` because it bypasses the decider path,
+- `TriggerTypeMention` contributes to soft load but not hard passive budget,
+- `intent_reply_threshold` and `intent_fallback_rate` come from config accessor rather than hardcoded constants,
+- active interjections cost less than passive replies.
+
+- [x] **Step 2: Run test to verify it fails**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/ratelimit ./internal/application/lark/messages/ops -run 'Test(SmartRateLimiter|Decider|ChatMsgOperator)' -count=1`
+Expected: FAIL because hard/soft budget separation and config-driven thresholds are not implemented.
+
+- [x] **Step 3: Implement minimal passive-reply ratelimit changes**
+
+Implement:
+- separate hard/soft weights,
+- willingness-aware score calculation in the decider,
+- config-driven thresholds/fallback rates,
+- chat operator mapping from intent reply mode to passive/interject trigger type.
+
+- [x] **Step 4: Run focused tests**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/ratelimit ./internal/application/lark/messages/ops -run 'Test(SmartRateLimiter|Decider|ChatMsgOperator)' -count=1`
+Expected: PASS.
+
+## Chunk 3: Prompt Split And Context Narrowing
+
+### Task 4: Split direct-response vs ambient-response prompt selection
+
+**Files:**
+- Modify: `internal/application/lark/agentruntime/chatflow/standard_plan.go`
+- Modify: `internal/application/lark/agentruntime/chatflow/standard_plan_test.go`
+- Modify: `internal/application/lark/handlers/chat_handler_test.go`
+- Modify: `internal/application/lark/messages/ops/reply_chat_op.go`
+- Modify: `internal/application/lark/messages/ops/chat_op.go`
+
+- [x] **Step 1: Write failing plan-selection tests**
+
+Add tests that verify:
+- direct replies choose the direct-response prompt path,
+- passive/interject replies choose the ambient prompt path,
+- passive prompt path no longer requires broad preloaded history to function.
+
+- [x] **Step 2: Run test to verify it fails**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/agentruntime/chatflow ./internal/application/lark/handlers -run 'Test(StandardPlan|GenerateChatSeq|ResolveChatExecutionMode)' -count=1`
+Expected: FAIL because prompt selection is still single-path.
+
+- [x] **Step 3: Implement minimal prompt/context split**
+
+Implement:
+- prompt variant selection in standard chat planning,
+- minimal default context for passive/direct replies,
+- continued use of `search_history` for broader recall instead of mandatory pre-spliced history blocks.
+
+- [x] **Step 4: Run focused tests**
+
+Run: `GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/agentruntime/chatflow ./internal/application/lark/handlers -run 'Test(StandardPlan|GenerateChatSeq|ResolveChatExecutionMode)' -count=1`
+Expected: PASS.
+
+## Chunk 4: Integrated Verification
+
+### Task 5: Run regression suite for touched domains
+
+**Files:**
+- Modify: `docs/superpowers/plans/2026-03-26-chat-passive-ratelimit-v2.md`
+
+- [x] **Step 1: Run the aggregated targeted suite**
+
+Run:
+`GOCACHE=/tmp/betago-gocache go test ./internal/application/lark/history ./internal/application/lark/handlers ./internal/application/lark/intent ./internal/application/lark/messages/ops ./internal/application/lark/ratelimit ./internal/application/lark/agentruntime/chatflow -count=1`
+
+Expected: PASS.
+
+- [x] **Step 2: Update plan checklist and capture follow-up debt**
+
+### Follow-up debt
+
+- `prompt_template_args.prompt_id=5` 仍然存在遗留模板耦合；本次通过“强约束系统提示 + 缩窄上下文”压住了问题，但下一轮最好迁到命名化 prompt config，而不是继续依赖数字 prompt ID。
+- `search_history` 已支持当前 chat 内的 metadata filter，但还没有进一步拆出 thread slice / recent slice 这类更低成本的专用读取工具。
+- 主索引现在作为 bot 自我身份判断的事实来源：bot 自己发出的消息保留真实 `user_id`，展示名仍可渲染成“你”；历史搜索和历史行渲染对旧数据里的 `"你"` sender alias 做兼容映射，暂不要求 retriever 同步这套身份元数据。
+- chat / agentic / continuation 三条 runtime user prompt 现在都会显式注入 `self_open_id` 与 `self_name`；`self_name` 通过带缓存的 `application/v6/applications/:app_id` 读取应用名，失败时回落到 `BaseInfo.RobotName`，便于模型在主索引历史和 mention 元数据里稳定识别“谁是自己”。
+- 主索引 mention 解析不再把 mention 直接吞掉；历史搜索结果会把 mention 还原成 `@姓名`，命中 bot 自己时渲染成 `@你`，便于模型直接理解“谁在@谁”。
+- 频控已经拆成 hard budget 和 soft load，但运营侧观测面板还看不到 direct/passive/interject 的拆分指标，后续应补 stats card 或 dashboard。
+- standard chat 现在默认不预取大段历史/主题摘要，后续可以继续根据 `needs_history / needs_web` 做更细粒度的 runtime 提示或自动 tool bias。
+
+Record any deferred work, especially:
+- optional new history tools for exact recent chat slices/thread slices,
+- migration from numeric prompt IDs to named prompt configs,
+- stats panel exposure for passive-vs-direct budgets.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add docs/superpowers/plans/2026-03-26-chat-passive-ratelimit-v2.md \
+  internal/application/lark/history/search.go \
+  internal/application/lark/history/search_test.go \
+  internal/application/lark/handlers/history_search_handler.go \
+  internal/application/lark/handlers/tools_test.go \
+  internal/application/lark/intentmeta/types.go \
+  internal/application/lark/intent/recognizer.go \
+  internal/application/lark/intent/recognizer_test.go \
+  internal/application/lark/messages/ops/chat_op.go \
+  internal/application/lark/messages/ops/intent_recognize_op_test.go \
+  internal/application/lark/ratelimit/integration.go \
+  internal/application/lark/ratelimit/rate_limiter.go \
+  internal/application/lark/ratelimit/rate_limiter_test.go \
+  internal/application/lark/agentruntime/chatflow/standard_plan.go \
+  internal/application/lark/agentruntime/chatflow/standard_plan_test.go \
+  internal/application/lark/handlers/chat_handler_test.go
+git commit -m "feat: add willingness-aware passive chat ratelimit"
+```