bobo-ai-cli 3.0.0 → 3.0.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (188) hide show
  1. package/bundled-skills/CORE_SKILLS.txt +18 -0
  2. package/dist/agent.js +25 -6
  3. package/dist/agent.js.map +1 -1
  4. package/dist/cli.js +9 -7
  5. package/dist/cli.js.map +1 -1
  6. package/dist/config.js +3 -2
  7. package/dist/config.js.map +1 -1
  8. package/dist/cost-tracker.js +1 -0
  9. package/dist/cost-tracker.js.map +1 -1
  10. package/dist/dream.js +6 -3
  11. package/dist/dream.js.map +1 -1
  12. package/dist/hooks.js +3 -2
  13. package/dist/hooks.js.map +1 -1
  14. package/dist/index.js +2 -2
  15. package/dist/index.js.map +1 -1
  16. package/dist/knowledge.js +3 -2
  17. package/dist/knowledge.js.map +1 -1
  18. package/dist/mcp-client.js +59 -41
  19. package/dist/mcp-client.js.map +1 -1
  20. package/dist/memory.js +24 -18
  21. package/dist/memory.js.map +1 -1
  22. package/dist/project.js +2 -1
  23. package/dist/project.js.map +1 -1
  24. package/dist/providers.js +1 -1
  25. package/dist/providers.js.map +1 -1
  26. package/dist/repl.js +4 -2
  27. package/dist/repl.js.map +1 -1
  28. package/dist/sessions.js +37 -4
  29. package/dist/sessions.js.map +1 -1
  30. package/dist/skill-router.js +2 -2
  31. package/dist/skill-router.js.map +1 -1
  32. package/dist/skills/loader.js +2 -2
  33. package/dist/skills/loader.js.map +1 -1
  34. package/dist/skills.js +9 -8
  35. package/dist/skills.js.map +1 -1
  36. package/dist/state/artifacts.js +4 -2
  37. package/dist/state/artifacts.js.map +1 -1
  38. package/dist/state/manager.js +4 -2
  39. package/dist/state/manager.js.map +1 -1
  40. package/dist/state/project-memory.js +4 -2
  41. package/dist/state/project-memory.js.map +1 -1
  42. package/dist/state/recovery.js +3 -2
  43. package/dist/state/recovery.js.map +1 -1
  44. package/dist/structured/loader.js +4 -2
  45. package/dist/structured/loader.js.map +1 -1
  46. package/dist/sub-agent-runner.js +3 -2
  47. package/dist/sub-agent-runner.js.map +1 -1
  48. package/dist/sub-agents.js +5 -3
  49. package/dist/sub-agents.js.map +1 -1
  50. package/dist/tool-governance.js +43 -12
  51. package/dist/tool-governance.js.map +1 -1
  52. package/dist/tools/advanced.js +3 -2
  53. package/dist/tools/advanced.js.map +1 -1
  54. package/dist/tools/browser.js +2 -2
  55. package/dist/tools/browser.js.map +1 -1
  56. package/dist/tools/claude-code.js +2 -2
  57. package/dist/tools/claude-code.js.map +1 -1
  58. package/dist/tools/index.js +36 -12
  59. package/dist/tools/index.js.map +1 -1
  60. package/dist/tools/process-manager.js +2 -2
  61. package/dist/tools/process-manager.js.map +1 -1
  62. package/dist/verification-agent.js +15 -10
  63. package/dist/verification-agent.js.map +1 -1
  64. package/dist/watcher.js +2 -2
  65. package/dist/watcher.js.map +1 -1
  66. package/dist/web.js +1 -1
  67. package/dist/web.js.map +1 -1
  68. package/dist/workflows/ask.js +1 -1
  69. package/dist/workflows/ask.js.map +1 -1
  70. package/package.json +17 -2
  71. package/bundled-skills/Skill_Seekers/SKILL.md +0 -1722
  72. package/bundled-skills/ab-test-setup/SKILL.md +0 -557
  73. package/bundled-skills/adversarial-verification/SKILL.md +0 -95
  74. package/bundled-skills/agent-sdk-dev/SKILL.md +0 -238
  75. package/bundled-skills/agent-tools/SKILL.md +0 -136
  76. package/bundled-skills/analytics-tracking/SKILL.md +0 -597
  77. package/bundled-skills/artifacts-builder/SKILL.md +0 -89
  78. package/bundled-skills/asana/SKILL.md +0 -13
  79. package/bundled-skills/brand-voice/SKILL.md +0 -481
  80. package/bundled-skills/browser-use/SKILL.md +0 -419
  81. package/bundled-skills/cache-optimization-skill/SKILL.md +0 -179
  82. package/bundled-skills/canvas-design/SKILL.md +0 -147
  83. package/bundled-skills/citation-validator/SKILL.md +0 -203
  84. package/bundled-skills/clangd-lsp/SKILL.md +0 -52
  85. package/bundled-skills/code-simplifier/SKILL.md +0 -13
  86. package/bundled-skills/commit-commands/SKILL.md +0 -258
  87. package/bundled-skills/competitor-alternatives/SKILL.md +0 -795
  88. package/bundled-skills/content-atomizer/SKILL.md +0 -910
  89. package/bundled-skills/content-research-writer/SKILL.md +0 -605
  90. package/bundled-skills/context7/SKILL.md +0 -13
  91. package/bundled-skills/copy-editing/SKILL.md +0 -494
  92. package/bundled-skills/copywriting/SKILL.md +0 -510
  93. package/bundled-skills/csharp-lsp/SKILL.md +0 -40
  94. package/bundled-skills/decision-making-framework/SKILL.md +0 -154
  95. package/bundled-skills/deep-research/SKILL.md +0 -236
  96. package/bundled-skills/developer-growth-analysis/SKILL.md +0 -335
  97. package/bundled-skills/direct-response-copy/SKILL.md +0 -2336
  98. package/bundled-skills/docker-expert/SKILL.md +0 -229
  99. package/bundled-skills/document-skills/SKILL.md +0 -13
  100. package/bundled-skills/documentation-expert/SKILL.md +0 -126
  101. package/bundled-skills/email-sequence/SKILL.md +0 -1061
  102. package/bundled-skills/email-sequences/SKILL.md +0 -910
  103. package/bundled-skills/example-plugin/SKILL.md +0 -72
  104. package/bundled-skills/explanatory-output-style/SKILL.md +0 -82
  105. package/bundled-skills/feature-dev/SKILL.md +0 -458
  106. package/bundled-skills/file-organizer/SKILL.md +0 -466
  107. package/bundled-skills/firebase.disabled/SKILL.md +0 -13
  108. package/bundled-skills/form-cro/SKILL.md +0 -488
  109. package/bundled-skills/free-tool-strategy/SKILL.md +0 -636
  110. package/bundled-skills/frontend-design/SKILL.md +0 -41
  111. package/bundled-skills/frontend-design-offical/SKILL.md +0 -55
  112. package/bundled-skills/gitlab/SKILL.md +0 -13
  113. package/bundled-skills/gopls-lsp/SKILL.md +0 -32
  114. package/bundled-skills/got-controller/SKILL.md +0 -218
  115. package/bundled-skills/greptile/SKILL.md +0 -72
  116. package/bundled-skills/hookify/SKILL.md +0 -376
  117. package/bundled-skills/image-editor/SKILL.md +0 -189
  118. package/bundled-skills/image-enhancer/SKILL.md +0 -109
  119. package/bundled-skills/jdtls-lsp/SKILL.md +0 -49
  120. package/bundled-skills/json-canvas/SKILL.md +0 -654
  121. package/bundled-skills/keyword-research/SKILL.md +0 -559
  122. package/bundled-skills/kotlin-lsp/SKILL.md +0 -28
  123. package/bundled-skills/laravel-boost/SKILL.md +0 -13
  124. package/bundled-skills/launch-strategy/SKILL.md +0 -394
  125. package/bundled-skills/lead-magnet/SKILL.md +0 -393
  126. package/bundled-skills/learning-output-style/SKILL.md +0 -106
  127. package/bundled-skills/linear/SKILL.md +0 -13
  128. package/bundled-skills/lua-lsp/SKILL.md +0 -47
  129. package/bundled-skills/marketing-ideas/SKILL.md +0 -720
  130. package/bundled-skills/marketing-psychology/SKILL.md +0 -534
  131. package/bundled-skills/mcp-builder/SKILL.md +0 -369
  132. package/bundled-skills/meeting-insights-analyzer/SKILL.md +0 -347
  133. package/bundled-skills/memory-evolution-system/SKILL.md +0 -172
  134. package/bundled-skills/multi-lens-thinking/SKILL.md +0 -407
  135. package/bundled-skills/nano-banana-pro/SKILL.md +0 -116
  136. package/bundled-skills/newsletter/SKILL.md +0 -736
  137. package/bundled-skills/notebooklm/SKILL.md +0 -296
  138. package/bundled-skills/obsidian-bases/SKILL.md +0 -634
  139. package/bundled-skills/obsidian-markdown/SKILL.md +0 -651
  140. package/bundled-skills/onboarding-cro/SKILL.md +0 -494
  141. package/bundled-skills/page-cro/SKILL.md +0 -379
  142. package/bundled-skills/paid-ads/SKILL.md +0 -624
  143. package/bundled-skills/paywall-upgrade-cro/SKILL.md +0 -651
  144. package/bundled-skills/php-lsp/SKILL.md +0 -36
  145. package/bundled-skills/playwright/SKILL.md +0 -13
  146. package/bundled-skills/plugin-dev/SKILL.md +0 -434
  147. package/bundled-skills/popup-cro/SKILL.md +0 -520
  148. package/bundled-skills/positioning-angles/SKILL.md +0 -330
  149. package/bundled-skills/pr-review-toolkit/SKILL.md +0 -359
  150. package/bundled-skills/pricing-strategy/SKILL.md +0 -777
  151. package/bundled-skills/proactive-self-improving/SKILL.md +0 -435
  152. package/bundled-skills/programmatic-seo/SKILL.md +0 -714
  153. package/bundled-skills/pyright-lsp/SKILL.md +0 -43
  154. package/bundled-skills/quality-assurance-framework/SKILL.md +0 -168
  155. package/bundled-skills/question-refiner/SKILL.md +0 -160
  156. package/bundled-skills/ralph-loop/SKILL.md +0 -205
  157. package/bundled-skills/refactoring-expert/SKILL.md +0 -103
  158. package/bundled-skills/referral-program/SKILL.md +0 -668
  159. package/bundled-skills/research-executor/SKILL.md +0 -164
  160. package/bundled-skills/review-with-security/SKILL.md +0 -12
  161. package/bundled-skills/rust-analyzer-lsp/SKILL.md +0 -50
  162. package/bundled-skills/schema-markup/SKILL.md +0 -647
  163. package/bundled-skills/security-audit-expert/SKILL.md +0 -124
  164. package/bundled-skills/security-expert/SKILL.md +0 -140
  165. package/bundled-skills/security-guidance/SKILL.md +0 -13
  166. package/bundled-skills/seedance-prompt/SKILL.md +0 -139
  167. package/bundled-skills/self-evolution/SKILL.md +0 -1160
  168. package/bundled-skills/seo-audit/SKILL.md +0 -432
  169. package/bundled-skills/seo-content/SKILL.md +0 -787
  170. package/bundled-skills/serena/SKILL.md +0 -13
  171. package/bundled-skills/signup-flow-cro/SKILL.md +0 -409
  172. package/bundled-skills/skill-manager/SKILL.md +0 -226
  173. package/bundled-skills/skill-share/SKILL.md +0 -98
  174. package/bundled-skills/slack/SKILL.md +0 -13
  175. package/bundled-skills/social-content/SKILL.md +0 -878
  176. package/bundled-skills/spec-flow-skill/SKILL.md +0 -124
  177. package/bundled-skills/stripe/SKILL.md +0 -13
  178. package/bundled-skills/supabase/SKILL.md +0 -13
  179. package/bundled-skills/swift-lsp/SKILL.md +0 -40
  180. package/bundled-skills/synthesizer/SKILL.md +0 -236
  181. package/bundled-skills/template-skill/SKILL.md +0 -16
  182. package/bundled-skills/theme-factory/SKILL.md +0 -72
  183. package/bundled-skills/tiktok-research/SKILL.md +0 -208
  184. package/bundled-skills/typescript-lsp/SKILL.md +0 -36
  185. package/bundled-skills/ui-ux-pro-max/SKILL.md +0 -247
  186. package/bundled-skills/visual-prompt-engineer/SKILL.md +0 -102
  187. package/bundled-skills/webapp-testing/SKILL.md +0 -111
  188. package/bundled-skills/wide-research/SKILL.md +0 -191
@@ -1,111 +0,0 @@
1
- ---
2
- id: "webapp-testing"
3
- title: "Web Application Testing"
4
- category: "dev-tools"
5
- tags: ["web application testing", "common pitfall", "best practices", "reference files"]
6
- triggers: []
7
- dependencies: []
8
- source: "E:/Bobo's Coding cache/.claude/skills/webapp-testing"
9
- ---
10
-
11
- ---
12
- name: webapp-testing
13
- description: Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
14
- license: Complete terms in LICENSE.txt
15
- ---
16
-
17
- # Web Application Testing
18
-
19
- To test local web applications, write native Python Playwright scripts.
20
-
21
- **Helper Scripts Available**:
22
-
23
- - `scripts/with_server.py` - Manages server lifecycle (supports multiple servers)
24
-
25
- **Always run scripts with `--help` first** to see usage. DO NOT read the source until you try running the script first and find that a customized solution is abslutely necessary. These scripts can be very large and thus pollute your context window. They exist to be called directly as black-box scripts rather than ingested into your context window.
26
-
27
- ## Decision Tree: Choosing Your Approach
28
-
29
- ```
30
- User task → Is it static HTML?
31
- ├─ Yes → Read HTML file directly to identify selectors
32
- │ ├─ Success → Write Playwright script using selectors
33
- │ └─ Fails/Incomplete → Treat as dynamic (below)
34
-
35
- └─ No (dynamic webapp) → Is the server already running?
36
- ├─ No → Run: python scripts/with_server.py --help
37
- │ Then use the helper + write simplified Playwright script
38
-
39
- └─ Yes → Reconnaissance-then-action:
40
- 1. Navigate and wait for networkidle
41
- 2. Take screenshot or inspect DOM
42
- 3. Identify selectors from rendered state
43
- 4. Execute actions with discovered selectors
44
- ```
45
-
46
- ## Example: Using with_server.py
47
-
48
- To start a server, run `--help` first, then use the helper:
49
-
50
- **Single server:**
51
-
52
- ```bash
53
- python scripts/with_server.py --server "npm run dev" --port 5173 -- python your_automation.py
54
- ```
55
-
56
- **Multiple servers (e.g., backend + frontend):**
57
-
58
- ```bash
59
- python scripts/with_server.py \
60
- --server "cd backend && python server.py" --port 3000 \
61
- --server "cd frontend && npm run dev" --port 5173 \
62
- -- python your_automation.py
63
- ```
64
-
65
- To create an automation script, include only Playwright logic (servers are managed automatically):
66
-
67
- ```python
68
- from playwright.sync_api import sync_playwright
69
-
70
- with sync_playwright() as p:
71
- browser = p.chromium.launch(headless=True) # Always launch chromium in headless mode
72
- page = browser.new_page()
73
- page.goto('http://localhost:5173') # Server already running and ready
74
- page.wait_for_load_state('networkidle') # CRITICAL: Wait for JS to execute
75
- # ... your automation logic
76
- browser.close()
77
- ```
78
-
79
- ## Reconnaissance-Then-Action Pattern
80
-
81
- 1. **Inspect rendered DOM**:
82
-
83
- ```python
84
- page.screenshot(path='/tmp/inspect.png', full_page=True)
85
- content = page.content()
86
- page.locator('button').all()
87
- ```
88
-
89
- 2. **Identify selectors** from inspection results
90
-
91
- 3. **Execute actions** using discovered selectors
92
-
93
- ## Common Pitfall
94
-
95
- ❌ **Don't** inspect the DOM before waiting for `networkidle` on dynamic apps
96
- ✅ **Do** wait for `page.wait_for_load_state('networkidle')` before inspection
97
-
98
- ## Best Practices
99
-
100
- - **Use bundled scripts as black boxes** - To accomplish a task, consider whether one of the scripts available in `scripts/` can help. These scripts handle common, complex workflows reliably without cluttering the context window. Use `--help` to see usage, then invoke directly.
101
- - Use `sync_playwright()` for synchronous scripts
102
- - Always close the browser when done
103
- - Use descriptive selectors: `text=`, `role=`, CSS selectors, or IDs
104
- - Add appropriate waits: `page.wait_for_selector()` or `page.wait_for_timeout()`
105
-
106
- ## Reference Files
107
-
108
- - **examples/** - Examples showing common patterns:
109
- - `element_discovery.py` - Discovering buttons, links, and inputs on a page
110
- - `static_html_automation.py` - Using file:// URLs for local HTML
111
- - `console_logging.py` - Capturing console logs during automation
@@ -1,191 +0,0 @@
1
- ---
2
- id: "wide-research"
3
- title: "Wide Research(Claude Code 版)多实例编排提示 / prompt.md"
4
- category: "research"
5
- tags: ["任务目标", "交付标准", "详细流程(claude code 语义)", "呈现与引用", "失败隔离、重试与覆盖率", "思考与写作准则(应用在每一轮输出)", "conclusion", "key evidence", "retrieval scope"]
6
- triggers: []
7
- dependencies: []
8
- source: "E:/Bobo's Coding cache/.claude/skills/wide-research"
9
- ---
10
-
11
- # Wide Research(Claude Code 版)多实例编排提示 / prompt.md
12
-
13
- 当用户在会话中提及 “Wide Research” 或引用此文件时,即表示你应加载本指令集。你是主控(Orchestrator),在 **Claude Code** 工作空间中编排可复用的多代理并行流程。任务可能涉及网页调研、代码检索、API 采样、数据清洗与成稿聚合。遵循安全与合规,优先最小化权限与最少外部调用。
14
-
15
- **重要:保持 Claude Code 的默认模型与底层配置不变;默认不开启 reasoning(或使用极低预算的 reasoning)。仅在得到用户明确授权时才提高推理档位与预算;且无论是否启用 reasoning,都不要在对外结果中暴露内部推理过程,只提供可验证的结论与证据。**
16
-
17
- ---
18
-
19
- ## 任务目标
20
-
21
- 第一步从用户的高层目标出发,推导需要并行处理的子目标集合(如主链接列表、数据集分片、模块清单)。
22
- 随后为每个子目标在 Claude Code 内部创建独立的“子代理执行单元”,控制权限与工具调用,并行产出结构化 Markdown。
23
- 将子结果按序聚合为统一成稿,对聚合结果做一次理智检查与最小化修复,返回最终 artefact 路径与要点摘要。
24
-
25
- ---
26
-
27
- ## 交付标准
28
-
29
- 成品是结构化、洞察驱动的整体,而不是子结果的机械拼接。子任务原文若需保留,单独写入 `aggregated_raw.md`,在成品中仅吸收关键洞察。
30
- 润色须逐段进行,每次修改都核对引用与上下文;交付前做双重体检:其一确认是分章节、多轮整合产物,而非一次性生成;其二评估细致程度,若单薄,则判断是素材不足还是统稿压缩过度,对应地补充调研或扩写润色。
31
-
32
- ---
33
-
34
- ## 详细流程(Claude Code 语义)
35
-
36
- 0. 预执行规划与摸底(必做)
37
- 主控亲自完成首轮摸底,明确目标、风险、约束与核心维度(主题簇、人物、地域、时间切片等)。若有公开目录/索引(标签页、API 列表),用最小化抓取取样;否则做案头调研拿到真实样本,并记录来源、时间与要点。
38
- 如工作区提供 Tavily MCP(或等价搜索 MCP),摸底阶段必须优先调用以获得至少一条直接相关样本;若不可用,记录原因并选择替代方案(例如轻量 `http(s)` 抓取)。
39
- 形成“草拟清单”:列出维度、已识别选项与样本、规模估算、缺口与不确定性。基于此给出可执行计划(子任务拆分、工具与脚本、输出格式、权限、软超时策略等),用用户语言汇报并待“执行/开始”的明确指令后继续。
40
-
41
- 1. 初始化与规划
42
- 在工作区新建不重复目录 `runs/<日期>-<任务摘要>-<随机后缀>/`,用于脚本、日志、子输出与聚合结果。明确目标、输出格式与评价标准。默认不开 reasoning;若获授权再以低预算开启。
43
-
44
- 2. 子目标识别
45
- 从素材或用户输入构造子目标列表。若源数据稀少,照实处理并记录原因;必要时由主控直接补位完成关键子目标。
46
-
47
- 3. 子目标并行执行(Claude Code 原生方式)
48
-
49
- **Claude Code 并行执行模型**:
50
- Claude Code 不支持生成独立子代理实例(不像多进程 CLI 工具),而是通过以下方式实现并行:
51
-
52
- - **单消息多工具调用**:在一条消息中同时调用多个 MCP 工具(如 `tavily_search`, `reddit__fetch_reddit_hot_threads`)
53
- - **分批队列执行**:将大量子目标分成多批,每批执行 4-6 个工具调用
54
- - **Task tool 代理**:对于大规模任务(>20 子目标),使用 Task tool 创建 Explore agents
55
-
56
- **并行执行决策树**:
57
-
58
- | 子目标数量 | 推荐方案 | 单批并行度 | 批次数 | 预期耗时 | 适用场景 |
59
- | ---------- | ------------------- | ------------- | ------ | ---------- | --------------------------- |
60
- | 1-5 | 单批 MCP 调用 | 4-6 个/批 | 1 批 | 2-5 分钟 | 快速调研、单一维度探索 |
61
- | 6-15 | 分批 MCP 调用 | 4-6 个/批 | 2-3 批 | 10-20 分钟 | 中等规模 subreddit/页面采样 |
62
- | 16-30 | Task tool (Explore) | 3-5 agents | 1-2 轮 | 20-40 分钟 | 大规模网页抓取、多维度研究 |
63
- | 30+ | Task tool + 分批 | 5-8 agents/批 | 2-4 批 | 40-90 分钟 | 行业全景扫描、竞品矩阵 |
64
-
65
- **执行约束**:
66
-
67
- - 单条消息 MCP 调用上限:**建议 4-6 个**(实测稳定)
68
- - 每个子目标工具交互轮次:**≤10 轮**
69
- - 软超时:小任务 5 分钟,大任务 15 分钟
70
-
71
- **输出规范**:
72
-
73
- - 每个子目标的结果写入 `child_outputs/<id>.md`
74
- - 原始数据缓存到 `raw/` 目录(避免重复请求)
75
- - 如需日志可写入 `logs/<id>.log`(可选)
76
-
77
- 所有联网需求优先使用 **MCP 搜索/抽取工具(如 tavily_search / tavily_extract)**;仅在无法使用 MCP 时才退回最小化的 `http(s)` 抓取。禁止过度并发、越权访问,遵守 robots 与版权。
78
-
79
- 4. 子代理产出规范(Markdown,自然语言)
80
- 先给出清晰结论与边界,再列关键证据,**就近附链接**(每个要点后给出来源),随后写明检索范围、调用轮次与缺口。遇到失败或超时,新增“失败原因/后续建议”小节,避免聚合阶段出现空白。
81
- 若涉及 PDF,优先使用可读性抽取工具;若工具不支持,再选择轻量抓取或手动要点化。
82
-
83
- 5. 并行执行与监控
84
- 记录每个子目标的开始/结束时间、耗时与状态。失败或超时的子任务要标记并在最终报告说明;如必要,对单个子任务重试,避免整体重跑。
85
- 遵循“缓存优先”,复用 `raw/` 中已存在的同源数据。
86
-
87
- 6. 数据聚合与报告生成
88
-
89
- **小规模任务(<10 child_outputs)- 手动聚合**:
90
- 逐个 Read `child_outputs/` 中的 Markdown 文件,提取关键信息并按预设结构组织。生成 `final_report.md`(精炼版)和 `aggregated_raw.md`(完整审计追溯)。在"基础稿"层面做去重、去噪,保留就近引用。
91
-
92
- **大规模任务(>10 child_outputs)- 结构化聚合**:
93
-
94
- 方案 A - 使用统一 Schema(推荐):在 child_outputs 生成阶段就使用标准化 YAML frontmatter,包含关键字段与数据置信度标签。聚合时可选择性使用 Python 脚本解析 frontmatter 并生成 CSV/JSON 汇总表。
95
-
96
- 方案 B - 分批聚合(当无法使用 Schema 时):每完成 5-10 个 child_outputs,立即生成"中间聚合稿"(如 `partial_report_01.md`)。最后做"总聚合",而非一次性处理所有文件。好处:避免 context window 爆炸,降低信息遗漏风险。
97
-
98
- 7. 结构设计与章纲
99
- 通读 `final_report.md` 与关键子输出,设计精修大纲 `polish_outline.md`:目标受众、章节顺序与每章核心论点、素材映射。
100
-
101
- 8. 分章润色与出稿
102
- 新建精修稿 `polished_report.md`,按大纲逐章撰写。每完成一章立即自查事实、引用与语言一致性;对重复信息、引用格式与待确认条目统一整理,同时保留核心事实与量化数据。避免一次性整篇重写,保持可追溯性。
103
-
104
- 9. 产出与交付
105
- 确认精修稿满足正式交付标准(结构完整、语气统一、引用准确)。对外交付以精修稿为准,并附路径与要点摘要;必要时补充后续跟进方式。不对外附带中间稿或内部笔记。
106
-
107
- ---
108
-
109
- ## 呈现与引用
110
-
111
- 在最终汇总中,每条要点后直接以 Markdown 链接形式附上来源,便于就地跳转查证,而不是把链接集中到文末。
112
- 对日期敏感的事实,优先写明**具体日期**(如 2025-10-23),避免“今天/最近/上周”导致歧义。
113
- 当多观点并存时,分别给出各自的高质量来源,并明示分歧点与证据等级。
114
-
115
- ---
116
-
117
- ## 失败隔离、重试与覆盖率
118
-
119
- 并行运行时若有子任务失败,先记录失败 ID 与日志,优先对单个子任务重试。重试前先检查 `child_outputs/<id>.md` 是否已存在合法结果,已存在则跳过,以节省配额并避免重复命中站点。
120
- 批量结束后,用轻量统计脚本或人工检查清单,确认缺失条目、空字段或引用缺口,确保在报告前被发现并补救。
121
-
122
- ---
123
-
124
- ## 思考与写作准则(应用在每一轮输出)
125
-
126
- 要有深度,有独立思考,给我惊喜(但是回答里别提惊喜)。
127
- 在行动前先判断:我为什么要问这个问题?背后隐藏的假设是什么?是否存在更优设定能带来更根本的解?
128
- 在动手前先定义**成功标准**:什么样的成品对用户真正有用,而非只是在单点上看起来“正确”。输出时紧扣这些标准。
129
- 开放性讨论时,以协作心态**扩展用户思考**,少抢答,多启发。用户未索取行动建议时不过度给出。
130
- 不要滥用顶层列表;尽量用自然语言解释清楚,减少术语。写作采用夹叙夹议、理性克制的语气;除直接引用外不要使用引号。
131
-
132
- ---
133
-
134
- ## Child Outputs 输出模板(`child_outputs/<id>.md` 格式规范)
135
-
136
- 每个子目标的研究结果应遵循以下结构:
137
-
138
- ```markdown
139
- # [子目标名称] Research Output
140
-
141
- ## Conclusion
142
-
143
- [一句话总结核心发现与适用边界]
144
-
145
- ## Key Evidence
146
-
147
- ### [维度 1]
148
-
149
- [要点化证据,每个要点后附来源链接]
150
-
151
- - 要点 1 ([source](url))
152
- - 要点 2 ([source](url))
153
-
154
- ### [维度 2]
155
-
156
- ...
157
-
158
- ## Retrieval Scope
159
-
160
- - **Tools used:** [列出使用的 MCP 工具]
161
- - **Timestamp:** [数据采集时间,如 2025-10-23]
162
- - **Coverage:** [检索范围说明]
163
-
164
- ## Gaps & Next Steps
165
-
166
- - [明确标注数据缺口]
167
- - [建议后续验证方向]
168
- ```
169
-
170
- **可选:数据置信度标注**(YAML frontmatter):
171
-
172
- ```yaml
173
- ---
174
- target: SubredditName
175
- data_confidence:
176
- field1: verified # 多源验证或官方来源
177
- field2: inferred # 基于间接证据推断
178
- field3: unverified # 缺少直接证据
179
- ---
180
- ```
181
-
182
- **执行约束**:
183
-
184
- - 工具优先级:优先 MCP(tavily_search / tavily_extract),不可用再最小化 `http(s)` 抓取;遵守 robots 与版权
185
- - 软超时与轮次:总工具轮次≤10;单子任务软超时 5–15 分钟,命中则收束并写清缺口与下一步
186
-
187
- ---
188
-
189
- ## 质检清单(交付前逐项自查)
190
-
191
- 结构完整、语气统一、引用准确;关键要点均有就近引用;日期与版本信息明确;分歧点已标注证据等级;聚合稿并非一次生成而是分章迭代;素材不足与压缩过度问题已区分并处理;最终给出 artefact 路径与摘要。