feat: categorized map layers, action buttons, file download, planner transfer fix

- Add 'categorized' layer type to MapPanel for per-feature polygon coloring
  based on property field + color mapping (DRL Opt_Type, Change_Type)
- Add category_labels for human-readable legend (耕地/林地/其他)
- Fix action buttons (export PDF etc.) using ChainlitAPI.callAction()
  instead of broken IAction.onClick() in @chainlit/react-client v0.3.1
- Fix DataPanel file click: non-CSV files now open/download via /api/user/files/
- Fix Planner not delegating to sub-agents: explicit transfer_to_agent()
  instructions in planner prompt (v7.1.2)
- Fix _final_map_update overwrite: layer-count comparison prevents
  summary-referenced old maps from replacing richer configs
- Fix PostGIS SRID detection: regex table extraction + ST_SRID fallback
- Clean FFI references from optimization/planner prompts and agent descriptions
- Add dynamic tool filtering (v7.5.6): intent-based tool category ContextVar
- Migrate genai import to google.genai unified SDK

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Gemini CLI

README.md

@@ -123,6 +123,7 @@ cd frontend && npm install && npm run dev
 | | MCP 工具市场 | 配置驱动的 MCP 服务器连接 + 工具聚合 + DB 持久化 + 管理 UI (v7.1) |
 | | 分析视角注入 | 用户自定义分析关注点，自动注入 Agent 提示词 (v7.1) |
 | | Memory ETL | 管道执行后自动提取关键发现，智能去重，配额管理 (v7.5) |
+| | 动态工具加载 | 按意图动态裁剪工具列表 (8 类别 + 10 核心工具)，ContextVar + ToolPredicate (v7.5) |
 | | 反思循环 | 全部 3 条管道含 LoopAgent 质量反思 (v7.1) |
 | **数据融合** | 融合引擎 (MMFE) | 五阶段流水线（画像→评估→对齐→融合→验证），10 种策略，5 种模态 |
 | | 语义匹配 | 五层渐进匹配：精确 → 等价组 → 嵌入相似度 → 单位感知 → 模糊 |
@@ -322,7 +323,7 @@ GitHub Actions 工作流（`.github/workflows/ci.yml`）在 push 到 `main`/`dev
 | v6.0 | 融合增强（栅格重投影、点云、流数据、语义增强、质量验证） | ✅ 完成 |
 | v7.0 | 向量嵌入匹配、LLM 策略路由、地理知识图谱、分布式计算 | ✅ 完成 |
 | v7.1 | MCP 管理 UI + DB 持久化、WorkflowEditor、分析视角注入、Prompt 版本管理、工具错误恢复、反思循环推广、端到端 Trace ID | ✅ 完成 |
-| v7.5 | Memory ETL 自动提取 ✅、Gemini Context Caching、动态工具加载、MCP 安全加固 + per-User 隔离 | 进行中 |
+| v7.5 | Memory ETL 自动提取 ✅、动态工具加载 ✅、Gemini Context Caching、MCP 安全加固 + per-User 隔离 | 进行中 |
 | v8.0 | DB 驱动自定义 Skills、RAG 知识库、DAG 工作流、失败学习与自适应、动态模型选择、评估门控 CI | 规划中 |
 | v9.0 | 实时协同编辑、边缘部署、数据连接器生态、多 Agent 并行、A2A 智能体互操作、主动探索与发现 | 远期 |
 

README_en.md

@@ -127,6 +127,7 @@ Default login: `admin` / `admin123` (seeded on first run). In-app self-registrat
 | | MCP Tool Market | Config-driven MCP server connection + tool aggregation + DB persistence + management UI (v7.1) |
 | | Analysis Perspective | User-defined analysis focus, auto-injected into agent prompts (v7.1) |
 | | Memory ETL | Auto-extract key findings after pipeline execution, smart dedup, quota management (v7.5) |
+| | Dynamic Tool Loading | Intent-based dynamic tool filtering (8 categories + 10 core tools), ContextVar + ToolPredicate (v7.5) |
 | | Reflection Loops | All 3 pipelines with LoopAgent quality reflection (v7.1) |
 | **Data Fusion** | Fusion Engine (MMFE) | Five-stage pipeline (Profile→Assess→Align→Fuse→Validate), 10 strategies, 5 modalities |
 | | Semantic Matching | Five-tier progressive: exact → equivalence groups → embedding similarity → unit-aware → fuzzy |
@@ -327,7 +328,7 @@ GitHub Actions workflow (`.github/workflows/ci.yml`) runs on push to `main`/`dev
 | v6.0 | Fusion Improvements (raster reprojection, point cloud, stream, quality) | ✅ Done |
 | v7.0 | Vector Embedding, LLM Strategy Routing, Knowledge Graph, Distributed Computing | ✅ Done |
 | v7.1 | MCP Management UI + DB Persistence, WorkflowEditor, Analysis Perspective, Prompt Versioning, Tool Error Recovery, Reflection Loop Expansion, End-to-End Trace ID | ✅ Done |
-| v7.5 | Memory ETL Auto-Extraction ✅, Gemini Context Caching, Dynamic Tool Loading, MCP Security + per-User Isolation | In Progress |
+| v7.5 | Memory ETL Auto-Extraction ✅, Dynamic Tool Loading ✅, Gemini Context Caching, MCP Security + per-User Isolation | In Progress |
 | v8.0 | DB-Driven Custom Skills, RAG Knowledge Base, DAG Workflow, Failure Learning & Adaptation, Dynamic Model Selection, Evaluation-Gated CI | Future |
 | v9.0 | Real-time Collaboration, Edge Deployment, Data Connectors, Multi-Agent Parallel, A2A Agent Interop, Proactive Exploration & Discovery | Long-term |
 

data_agent/agent.py

@@ -79,6 +79,7 @@
 from .spatial_statistics import spatial_autocorrelation, local_moran, hotspot_analysis
 from .gis_processors import _generate_output_path, _resolve_path
 from .database_tools import T_TABLE_OWNERSHIP
+from .tool_filter import intent_tool_predicate
 
 # ---------------------------------------------------------------------------
 # Tool filter presets — reusable across agents
@@ -124,8 +125,8 @@
 
 data_exploration_agent = LlmAgent(
     name="DataExploration",
-    instruction=get_prompt("optimization", "data_exploration_agent_instruction"),
-    description="数据质量审计与治理专家",
+    instruction=get_prompt("optimization", "data_exploration_opt_instruction"),
+    description="优化管道数据准备专家",
     model=MODEL_STANDARD,
     output_key="data_profile",
     after_tool_callback=_self_correction_after_tool,
@@ -138,8 +139,8 @@
 
 data_processing_agent = LlmAgent(
     name="DataProcessing",
-    instruction=get_prompt("optimization", "data_processing_agent_instruction"),
-    description="特征工程与预处理专家",
+    instruction=get_prompt("optimization", "data_processing_opt_instruction"),
+    description="优化管道数据预处理专家",
     model=MODEL_STANDARD,
     output_key="processed_data",
     after_tool_callback=_self_correction_after_tool,
@@ -178,7 +179,7 @@
 quality_checker_agent = LlmAgent(
     name="QualityChecker",
     instruction=get_prompt("optimization", "quality_checker_instruction"),
-    description="分析结果质量审查员。验证FFI/DRL/遥感指标合理性。",
+    description="分析结果质量审查员。验证DRL优化/遥感指标合理性。",
     model=MODEL_FAST,
     output_key="quality_verdict",
     tools=[approve_quality],
@@ -301,21 +302,21 @@
     after_tool_callback=_self_correction_after_tool,
     tools=[
         ExplorationToolset(tool_filter=_TRANSFORM_TOOLS),
-        GeoProcessingToolset(),
-        LocationToolset(),
+        GeoProcessingToolset(tool_filter=intent_tool_predicate),
+        LocationToolset(tool_filter=intent_tool_predicate),
         DatabaseToolset(tool_filter=_DB_READ_DESCRIBE + ["share_table", "import_to_postgis"]),
         FileToolset(),
         MemoryToolset(),
-        AdminToolset(),
-        RemoteSensingToolset(),
-        SpatialStatisticsToolset(),
-        SemanticLayerToolset(),
-        StreamingToolset(),
-        TeamToolset(),
-        DataLakeToolset(),
+        AdminToolset(tool_filter=intent_tool_predicate),
+        RemoteSensingToolset(tool_filter=intent_tool_predicate),
+        SpatialStatisticsToolset(tool_filter=intent_tool_predicate),
+        SemanticLayerToolset(tool_filter=intent_tool_predicate),
+        StreamingToolset(tool_filter=intent_tool_predicate),
+        TeamToolset(tool_filter=intent_tool_predicate),
+        DataLakeToolset(tool_filter=intent_tool_predicate),
         McpHubToolset(pipeline="general"),
-        FusionToolset(),
-        KnowledgeGraphToolset(),
+        FusionToolset(tool_filter=intent_tool_predicate),
+        KnowledgeGraphToolset(tool_filter=intent_tool_predicate),
     ] + _arcpy_tools,
 )
 
@@ -411,15 +412,15 @@ def _make_planner_processor(name: str, **overrides) -> LlmAgent:
         after_tool_callback=_self_correction_after_tool,
         tools=[
             ExplorationToolset(tool_filter=_TRANSFORM_TOOLS),
-            GeoProcessingToolset(),
-            LocationToolset(),
+            GeoProcessingToolset(tool_filter=intent_tool_predicate),
+            LocationToolset(tool_filter=intent_tool_predicate),
             RemoteSensingToolset(tool_filter=["describe_raster", "download_lulc", "download_dem"]),
-            StreamingToolset(),
+            StreamingToolset(tool_filter=intent_tool_predicate),
             DataLakeToolset(tool_filter=_DATALAKE_READ),
             DatabaseToolset(tool_filter=["import_to_postgis"]),
             McpHubToolset(pipeline="planner"),
-            FusionToolset(),
-            KnowledgeGraphToolset(),
+            FusionToolset(tool_filter=intent_tool_predicate),
+            KnowledgeGraphToolset(tool_filter=intent_tool_predicate),
         ] + _arcpy_tools,
     )
     defaults.update(overrides)
@@ -431,7 +432,7 @@ def _make_planner_analyzer(name: str, **overrides) -> LlmAgent:
     defaults = dict(
         name=name,
         instruction=get_prompt("planner", "planner_analyzer_instruction"),
-        description="FFI破碎化指数、DRL深度强化学习布局优化、遥感分析、空间统计专家。",
+        description="DRL深度强化学习布局优化、遥感分析、空间统计专家。",
         model=MODEL_STANDARD,
         output_key="analysis_report",
         disallow_transfer_to_peers=True,
@@ -448,7 +449,7 @@ def _make_planner_visualizer(name: str, **overrides) -> LlmAgent:
         name=name,
         instruction=get_prompt("planner", "planner_visualizer_instruction"),
         description="地理空间可视化专家。交互地图、Choropleth、热力图、气泡图、PNG导出。",
-        model=MODEL_FAST,
+        model=MODEL_STANDARD,
         output_key="visualizations",
         disallow_transfer_to_peers=True,
         tools=[
@@ -492,7 +493,7 @@ def _make_planner_visualizer(name: str, **overrides) -> LlmAgent:
 
 analyze_viz_workflow = SequentialAgent(
     name="AnalyzeAndVisualize",
-    description="分析→可视化 一体化工作流。执行FFI/DRL/统计分析后自动生成可视化。",
+    description="分析→可视化 一体化工作流。执行DRL/统计分析后自动生成可视化。",
     sub_agents=[
         _make_planner_analyzer("WFAnalyzer"),
         _make_planner_visualizer("WFVisualizer"),