feat: 研究质量改进、谛听品牌与深空 Demo UI

.env.compose.example

@@ -1,5 +1,5 @@
 # ===============================
-# OpenClaw-MiroSearch Compose 环境变量示例
+# 谛听 Compose 环境变量示例
 # 使用方法：
 # cp .env.compose.example .env.compose
 # docker compose --env-file .env.compose up -d --build
@@ -60,6 +60,8 @@ DEFAULT_VERIFICATION_MIN_SEARCH_ROUNDS=3
 #          页面刷新 / 断电后重连 SSE 续看进度。
 #          需配套 api/worker/valkey 服务（compose 已默认包含）。
 BACKEND_MODE=api
+# 生产/Compose 关闭静态热读，允许短缓存（本地开发默认不设或设 1）
+DITING_STATIC_HOT=0
 # host-network 模式下使用本机 API 端口；端口映射模式可写 http://api:8090
 API_BASE_URL=http://127.0.0.1:8090
 # API 模式的 Bearer Token 在上方“API 鉴权”段统一配置

.gitignore

@@ -221,6 +221,7 @@ data/
 .idea/
 .windsurf/
 .claude/
+.workbuddy-ai/
 CLAUDE.md
 
 .DS_Store

README.md

@@ -1,8 +1,12 @@
-# OpenClaw-MiroSearch
+# Diting (谛听)
+
+<p align="center">
+  <img src="assets/diting_logo.png" alt="Diting Logo" width="320" />
+</p>
 
 [English](README.md) | [中文](README_zh.md)
 
-OpenClaw-MiroSearch is an agentic research service built on MiroThinker. It combines multi-provider web search, full-page extraction, multi-step reasoning, source verification, asynchronous execution, and structured Markdown reports.
+Diting (谛听) is an open-source agentic research service built on MiroThinker. It combines multi-provider web search, full-page extraction, multi-step reasoning, source verification, asynchronous execution, and structured Markdown reports.
 
 Current stable release: **v0.2.11**. See the [Changelog](docs/CHANGELOG.md) for released changes and the [Roadmap](docs/ROADMAP.md) for planned work.
 
@@ -104,12 +108,115 @@ These are client recommendations, not immutable server defaults. Omitted optiona
 - [Changelog](docs/CHANGELOG.md)
 - [Security](docs/SECURITY.md)
 - [Contributing](docs/CONTRIBUTING.md)
-- [OpenClaw skill](skills/openclaw-mirosearch/SKILL.md)
+- [Diting skill](skills/diting/SKILL.md)
 
 ## Development
 
 Python 3.12+ is required for the primary applications and library.
 
+```bash
+# Install app dependencies
+cd apps/gradio-demo && uv sync
+cd ../miroflow-agent && uv sync
+cd ../../libs/miroflow-tools && uv sync
+```
+
+If you need legacy compatibility or want to reuse the Demo UI directly, Gradio API remains available:
+
+```bash
+BASE_URL="http://127.0.0.1:8080"
+curl -sS "$BASE_URL/gradio_api/info"
+```
+
+## For OpenClaw / AI Agents
+
+Project positioning:
+
+- Provides web research capability callable by upper-layer agents
+- Supports four-dimensional control: mode, routing, search depth, and output detail
+- Uses SSE terminal events so agents can determine task completion
+
+Recommended agent calling loop:
+
+1. Call `GET /health` for health check
+1. Submit `POST /v1/research`
+1. Poll `GET /v1/research/{task_id}` or subscribe to `GET /v1/research/{task_id}/stream`
+1. When `status=completed` or `cached`, consume only the final Markdown
+
+Skill guidance:
+
+- Simple search, single-fact lookup, and cost-first usage: use the repository-distributed `searxng` skill
+  - Repository: `skills/searxng/`
+  - Packaged file: `skills/searxng.zip`
+- Deep research or high-quality retrieval: use the `diting` skill
+  - Skill docs: [`skills/diting/SKILL.md`](skills/diting/SKILL.md)
+  - Usage docs: [`skills/diting/references/usage.md`](skills/diting/references/usage.md)
+
+Skill acquisition and installation:
+
+- Recommended dual-skill bundle: `skills/openclaw-search-skills-bundle.zip`
+- Simple search skill: `skills/searxng/`
+- Repository: `skills/diting/`
+- Packaged file: `skills/diting.zip`
+- Installation guide: [`skills/diting/references/skill-install.md`](skills/diting/references/skill-install.md)
+- API docs: [`skills/diting/references/api.md`](skills/diting/references/api.md)
+- AI Agent integration: [`docs/API_SPEC.md`](docs/API_SPEC.md)
+
+## Recommended Configuration Baseline
+
+- **Default production**: `mode=balanced` + `search_profile=parallel-trusted`
+- **High-risk fact-checking**: `mode=verified` + `search_profile=parallel-trusted`
+- **Quota-priority**: `mode=quota` + `search_profile=searxng-only`
+- **Verification depth**: `search_result_num=30` + `verification_min_search_rounds=4`
+
+> For the full list of routing environment variables, see [`apps/miroflow-agent/README.md`](apps/miroflow-agent/README.md#4-检索路由配置) and [`docs/API_SPEC.md`](docs/API_SPEC.md)
+
+## Changelog
+
+- Release `0.2.4` highlights:
+  - `scrape_url` now supports PDF extraction with a 20MB streamed body limit
+  - JSON / RSS / Atom / XML payloads can pass through with structured fields (`json_keys`, `feed_title`, `entries`, `xml_root`)
+  - Redirect handling now uses streamed responses and closes intermediate 30x hops eagerly
+  - Local Docker end-to-end verification passed on the `app + api + worker + searxng + valkey` stack
+  - See [`docs/SCRAPING_ITERATION_PLAN.md`](docs/SCRAPING_ITERATION_PLAN.md) for the full T1–T9 scraping roadmap
+- Release `0.2.2` highlights:
+  - API-mode regression fix: `mode` / `search_profile` / `search_result_num` / `verification_min_search_rounds` / `output_detail_level` are now respected end-to-end via `services/profile_resolver.py`
+  - Demo crash-recovery: `BACKEND_MODE=api` plus `?task_id=xxx` URL bridge — refresh / disconnect resumes the same task via SSE replay
+  - MCP tool `scrape_url`: lightweight `httpx + BeautifulSoup` scraper with SSRF guard so the LLM can "open the page" when `google_search` snippets are insufficient
+  - Worker cancel watcher hardened against Redis hiccups; unresponsive pipelines are abandoned after a 10s timeout
+  - Dockerfile uses a domestic apt mirror by default; compose builds run with `network: host`; `scripts/deploy/build_images.sh` bypasses BuildKit's `network.host` entitlement prompt
+- Release `0.2.1` highlights:
+  - Clickable `[N]` references in research summaries pointing to the report's References / 参考文献 section
+  - `api-worker` startup command pinned to `.venv/bin/python` for reliable arq worker boot
+- Release `0.2.0` highlights:
+  - Async task queue (arq + Valkey), persistent SSE event streams, cache and metadata persistence
+  - `SearchProvider` Protocol + `ProviderRegistry` (Serper / SerpAPI / SearXNG)
+- Full history: [`docs/CHANGELOG.md`](docs/CHANGELOG.md)
+
+## Documentation Index
+
+- Overview: [`docs/README.md`](docs/README.md)
+- Architecture: [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md)
+- Deployment: [`docs/DEPLOY.md`](docs/DEPLOY.md)
+- API spec & Agent integration: [`docs/API_SPEC.md`](docs/API_SPEC.md)
+- Roadmap: [`docs/ROADMAP.md`](docs/ROADMAP.md)
+- Changelog: [`docs/CHANGELOG.md`](docs/CHANGELOG.md)
+- Demo docs: [`apps/gradio-demo/README.md`](apps/gradio-demo/README.md)
+- API server docs: [`apps/api-server/README.md`](apps/api-server/README.md)
+- Agent docs: [`apps/miroflow-agent/README.md`](apps/miroflow-agent/README.md)
+- Tools docs: [`libs/miroflow-tools/README.md`](libs/miroflow-tools/README.md)
+- Diting skill package: [`skills/diting/SKILL.md`](skills/diting/SKILL.md)
+
+## Open Source Collaboration
+
+- Contributing, governance, support & release: [`docs/CONTRIBUTING.md`](docs/CONTRIBUTING.md)
+- Security policy: [`docs/SECURITY.md`](docs/SECURITY.md)
+- Code of conduct: [`docs/CODE_OF_CONDUCT.md`](docs/CODE_OF_CONDUCT.md)
+- Changelog: [`docs/CHANGELOG.md`](docs/CHANGELOG.md)
+
+
+## Development Validation
+
 ```bash
 # Repository-wide checks
 just lint

README_zh.md

@@ -1,8 +1,12 @@
-# OpenClaw-MiroSearch
+# 谛听（Diting）
+
+<p align="center">
+  <img src="assets/diting_logo.png" alt="Diting Logo" width="320" />
+</p>
 
 [English](README.md) | [中文](README_zh.md)
 
-OpenClaw-MiroSearch 是基于 MiroThinker 构建的智能体研究服务，集成多源网页检索、正文抓取、多步推理、来源校验、异步执行和结构化 Markdown 报告。
+谛听（Diting）是基于 MiroThinker 构建的开源智能体研究服务，集成多源网页检索、正文抓取、多步推理、来源校验、异步执行和结构化 Markdown 报告。
 
 当前稳定版本：**v0.2.11**。已发布变更见 [变更记录](docs/CHANGELOG_zh.md)，未来计划见 [路线图](docs/ROADMAP_zh.md)。
 
@@ -37,6 +41,8 @@ OpenClaw-MiroSearch 是基于 MiroThinker 构建的智能体研究服务，集
 
 需要 Docker Compose v2、一个 OpenAI 兼容或 Anthropic LLM 端点，以及至少一个可用搜索来源。
 
+<p align="center"><img src="assets/demo-screenshot.png" alt="Demo Screenshot" width="900" /></p>
+
 ```bash
 cp .env.compose.example .env.compose
 # 编辑 .env.compose，替换 LLM 和搜索服务的占位凭据。
@@ -104,20 +110,116 @@ curl -sS -N http://127.0.0.1:8090/v1/research/<task_id>/stream
 - [变更记录](docs/CHANGELOG_zh.md)
 - [安全策略](docs/SECURITY_zh.md)
 - [贡献指南](docs/CONTRIBUTING_zh.md)
-- [OpenClaw Skill](skills/openclaw-mirosearch/SKILL.md)
+- [Diting Skill](skills/diting/SKILL.md)
 
 ## 开发
 
 主要应用和工具库要求 Python 3.12+。
 
+```bash
+# 安装依赖
+cd apps/gradio-demo && uv sync
+cd ../miroflow-agent && uv sync
+cd ../../libs/miroflow-tools && uv sync
+```
+
+如需兼容旧链路或直接复用 Demo UI，仍可使用 Gradio API：
+
+```bash
+BASE_URL="http://127.0.0.1:8080"
+curl -sS "$BASE_URL/gradio_api/info"
+```
+
+## 面向 OpenClaw / AI Agent
+
+这个项目的定位：
+
+- 提供可被上层智能体调用的联网研究能力
+- 支持模式、路由、检索深度与输出篇幅四维可控
+- 通过 SSE 终态事件，保证智能体编排时可判断任务完成
+
+推荐给 AI Agent 的调用闭环：
+
+1. 先调 `GET /health` 探活
+1. 发起 `POST /v1/research`
+1. 轮询 `GET /v1/research/{task_id}` 或订阅 `GET /v1/research/{task_id}/stream`
+1. `status=completed` 或 `cached` 时，只消费最终 Markdown
+
+Skill 使用建议（先分流）：
+
+- 简单搜索（快速网页检索、单事实查询）：优先使用仓库内分发的 `searxng` skill
+  - 仓库目录：`skills/searxng/`
+  - 打包文件：`skills/searxng.zip`
+- 深度检索或高质量检索（多来源交叉、核查、研究报告）：使用 `diting` skill
+
+Skill 安装：
+
+- 推荐双 skill 打包：`skills/openclaw-search-skills-bundle.zip`
+- 简单搜索 skill：`skills/searxng/`
+- 仓库目录：`skills/diting/`
+- 打包文件：`skills/diting.zip`
+- 安装说明：[`skills/diting/references/skill-install.md`](skills/diting/references/skill-install.md)
+
+Skill 使用：
+
+- 使用说明：[`skills/diting/references/usage.md`](skills/diting/references/usage.md)
+- API 说明：[`skills/diting/references/api.md`](skills/diting/references/api.md)
+- AI Agent 接入详解：[`docs/API_SPEC.md`](docs/API_SPEC.md)
+
+## 建议配置基线
+
+- **默认生产**：`mode=balanced` + `search_profile=parallel-trusted`
+- **高风险事实核查**：`mode=verified` + `search_profile=parallel-trusted`
+- **额度优先**：`mode=quota` + `search_profile=searxng-only`
+- **核查深度**：`search_result_num=30` + `verification_min_search_rounds=4`
+
+> 完整路由环境变量说明请参见 [`apps/miroflow-agent/README.md`](apps/miroflow-agent/README.md) 和 [`docs/API_SPEC.md`](docs/API_SPEC.md)
+
+## 版本亮点
+
+- `0.2.4` 版本亮点：
+  - `scrape_url` 已支持 PDF 抽取，并带 20MB 流式响应体上限
+  - 已支持 JSON / RSS / Atom / XML 结构化直通，返回 `json_keys`、`feed_title`、`entries`、`xml_root` 等字段
+  - 重定向链路改为流式响应，并在中间 30x hop 及时关闭连接
+  - 本地 Docker `app + api + worker + searxng + valkey` 真实端到端验证已通过
+- `0.2.2` 版本亮点：
+  - API 模式严重回归修复：`mode` / `search_profile` / `search_result_num` / `verification_min_search_rounds` / `output_detail_level` 已可端到端透传
+  - Demo 断线重连：`BACKEND_MODE=api` 配合 `?task_id=xxx` 可通过 SSE 回放恢复任务
+  - MCP `scrape_url` 初版上线：基于 `httpx + BeautifulSoup`，在 `google_search` 摘要不足时让 LLM 主动打开正文
+  - 详细抓取路线图见 [`docs/SCRAPING_ITERATION_PLAN.md`](docs/SCRAPING_ITERATION_PLAN.md)
+
+## 文档索引
+
+- 文档总览：[`docs/README.md`](docs/README.md)
+- 架构概览：[`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md)
+- 部署指南：[`docs/DEPLOY.md`](docs/DEPLOY.md)
+- API 规格 & Agent 接入：[`docs/API_SPEC.md`](docs/API_SPEC.md)
+- 路线图：[`docs/ROADMAP.md`](docs/ROADMAP.md)
+- 抓取能力迭代计划：[`docs/SCRAPING_ITERATION_PLAN.md`](docs/SCRAPING_ITERATION_PLAN.md)（T1-T9，对应 v0.2.4 → v0.3.0）
+- 变更记录：[`docs/CHANGELOG.md`](docs/CHANGELOG.md)
+- Demo 说明：[`apps/gradio-demo/README.md`](apps/gradio-demo/README.md)
+- API Server 说明：[`apps/api-server/README.md`](apps/api-server/README.md)
+- Agent 说明：[`apps/miroflow-agent/README.md`](apps/miroflow-agent/README.md)
+- 工具层说明：[`libs/miroflow-tools/README.md`](libs/miroflow-tools/README.md)
+- Diting 技能包：[`skills/diting/SKILL.md`](skills/diting/SKILL.md)
+
+## 开源协作文档
+
+- 贡献、治理、支持与发布：[`docs/CONTRIBUTING.md`](docs/CONTRIBUTING.md)
+- 安全策略：[`docs/SECURITY.md`](docs/SECURITY.md)
+- 行为准则：[`docs/CODE_OF_CONDUCT.md`](docs/CODE_OF_CONDUCT.md)
+- 变更记录：[`docs/CHANGELOG.md`](docs/CHANGELOG.md)
+
+## 开发验证
+
 ```bash
 # 仓库级检查
 just lint
 just sort-imports
 just format
 just format-md
 
-# 各模块测试
+# 分模块测试
 cd apps/miroflow-agent && uv sync && uv run pytest
 cd apps/api-server && uv sync && AUTH_DISABLED=1 uv run pytest
 cd apps/gradio-demo && uv sync && uv run pytest

apps/api-server/README.md

@@ -1,4 +1,4 @@
-# MiroSearch API Server
+# Diting API Server
 
 [English](README.md) | [中文](README_zh.md)
 

apps/api-server/main.py

@@ -78,8 +78,8 @@ async def _cleanup_loop():
 
 
 app = FastAPI(
-    title="MiroSearch API",
-    description="OpenClaw-MiroSearch 标准 HTTP API，独立于 Gradio Demo",
+    title="Diting API",
+    description="谛听标准 HTTP API，独立于 Gradio Demo",
     version=settings.api_version,
     docs_url="/docs",
     redoc_url="/redoc",

apps/api-server/models.py

@@ -29,6 +29,9 @@
 MAX_VERIFICATION_SEARCH_ROUNDS = 8
 
 
+VALID_RESEARCH_INTENSITIES = ("light", "standard", "deep")
+
+
 class ResearchRequest(BaseModel):
     """POST /v1/research 请求体。
 
@@ -55,6 +58,10 @@ class ResearchRequest(BaseModel):
         default=None,
         description="输出篇幅档位",
     )
+    research_intensity: Optional[str] = Field(
+        default=None,
+        description="研究强度：light / standard / deep（可选，默认 standard）",
+    )
     caller_id: Optional[str] = Field(
         default=None, description="调用方 ID，用于定向取消"
     )
@@ -114,6 +121,18 @@ def validate_output_detail_level(cls, v: Optional[str]) -> Optional[str]:
             )
         return normalized
 
+    @field_validator("research_intensity")
+    @classmethod
+    def validate_research_intensity(cls, v: Optional[str]) -> Optional[str]:
+        if v is None:
+            return None
+        normalized = v.strip().lower()
+        if normalized not in VALID_RESEARCH_INTENSITIES:
+            raise ValueError(
+                f"research_intensity must be one of {VALID_RESEARCH_INTENSITIES}, got '{v}'"
+            )
+        return normalized
+
     @field_validator("caller_id")
     @classmethod
     def validate_caller_id(cls, v: Optional[str]) -> Optional[str]:
@@ -163,6 +182,17 @@ class ErrorResponse(BaseModel):
 # ---- 新增任务状态模型 ----
 
 
+class EffectiveConfig(BaseModel):
+    """实际生效的配置（记录任务真正使用的参数）。"""
+
+    mode: str
+    search_profile: str
+    search_result_num: int
+    verification_min_search_rounds: int
+    output_detail_level: str
+    research_intensity: Optional[str] = None
+
+
 class ResearchTaskMeta(BaseModel):
     """任务元数据。"""
 
@@ -175,6 +205,8 @@ class ResearchTaskMeta(BaseModel):
     search_result_num: int = 20
     verification_min_search_rounds: int = 3
     output_detail_level: str = "detailed"
+    research_intensity: Optional[str] = None
+    effective_config: Optional[EffectiveConfig] = None
     created_at: float = 0.0
     started_at: Optional[float] = None
     finished_at: Optional[float] = None

apps/api-server/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "api-server"
 version = "0.2.0"
-description = "FastAPI API server for MiroSearch, independent from Gradio demo"
+description = "FastAPI API server for Diting, independent from Gradio demo"
 readme = "README.md"
 requires-python = ">=3.12"
 dependencies = [