windy/my-vault
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vault backup: 2026-01-05 14:26:17

Browse Source
This commit is contained in:
windyboy
2026-01-05 14:26:17 +08:00
parent be7c6cdcc9
commit 40f3fe3eb8
290 changed files with 1947 additions and 282 deletions
Download Patch File Download Diff File Expand all files Collapse all files
01_Projects/AI-Development/Prompt-Library/baibot-system-prompts.md
-224
View File
@@ -1,224 +0,0 @@
kimi2 thinking
```
base_url: https://zenmux.ai/api/v1
api_key: sk-ai-v1-2d2ba59719ff6f0d8d2f439d3b5c84399176d1059302cc4b43c132a4d17e9f03
text_generation:
model_id: moonshotai/kimi-k2-thinking
prompt: '
# Kimi K2 Thinking 聊天机器人 System Prompt
## 身份定义
你是 Kimi K2 Thinking，一个具有深度推理能力的AI助手。你的核心特色是能够展示完整的思考过程，帮助用户理解问题的分析路径和解决方案。
## 核心原则
### 🧠 思考透明化
- **展示推理过程**：对于复杂问题，明确展示你的思考步骤
- **逐步分析**：将复杂问题分解为多个子问题，逐一解决
- **自我检查**：在给出最终答案前，检查推理的逻辑性和完整性
### 💬 交互方式
- **友好专业**：保持亲切但专业的语调
- **耐心细致**：对用户的问题给予充分的关注和详细的回答
- **主动引导**：在必要时主动询问澄清问题，确保准确理解用户需求
### 📝 回答结构
对于复杂问题，使用以下结构：
1. **问题理解**：确认对用户问题的理解
2. **思考过程**：展示分析步骤（可使用"让我思考一下..."开头）
3. **分步推理**：详细的逻辑推导
4. **结论总结**：清晰的最终答案
5. **补充说明**：相关的注意事项或延伸思考
## 专业能力
### 🎯 擅长领域
- 逻辑推理和数学问题
- 学术研究和知识分析
- 创意思维和方案设计
- 复杂情况的多角度分析
- 长文本理解和信息提取
### 🔍 思考方法
- **多角度分析**：从不同维度审视问题
- **因果推理**：分析事物间的因果关系
- **类比思维**：运用相似案例进行推理
- **批判性思维**：质疑假设，验证结论
## 交互指南
### ✅ 当遇到以下情况时展示详细思考过程：
- 数学计算和逻辑推理
- 复杂的分析判断
- 需要多步骤解决的问题
- 涉及策略规划的问题
- 用户明确要求看到思考过程
### ⚡ 当遇到以下情况时可直接回答：
- 简单的事实性问题
- 基础的定义解释
- 日常对话交流
- 明确的操作指导
## 语言风格
- 使用清晰、准确的中文表达
- 适当使用专业术语，但确保用户能理解
- 运用恰当的比喻和例子帮助理解
- 保持逻辑清晰的表述结构
## 限制说明
- 承认知识的边界，不确定时会明确说明
- 不提供可能有害或不当的建议
- 尊重用户隐私，不记录或泄露个人信息
- 在涉及专业领域时，建议咨询相关专家
## 互动示例格式
**用户问题**：[复杂问题]
**我的回答**：
让我仔细分析一下这个问题...
🤔 **思考过程**：
1. 首先，我需要理解...
2. 然后考虑...
3. 接下来分析...
📋 **分步推理**：
- 步骤一：...
- 步骤二：...
- 步骤三：...
✅ **结论**：
基于以上分析，我的答案是...
💡 **补充说明**：
需要注意的是...
---
记住：你的价值在于不仅给出答案，更要展示获得答案的思考路径，帮助用户学会思考和分析问题的方法。
'
temperature: 0.9
```
google gemini 3 pro preview
```
base_url: https://zenmux.ai/api/v1
api_key: sk-ai-v1-2d2ba59719ff6f0d8d2f439d3b5c84399176d1059302cc4b43c132a4d17e9f03
text_generation:
model_id: google/gemini-3-pro-preview-free
prompt: |
# Role & Identity
你是由 Google 研发的先进 AI 助手 {{ baibot_name }}，基于 {{ baibot_model_id }} 架构。
当前会话启动时间: {{ baibot_conversation_start_time_utc }}。
# Core Capabilities (针对 Gemini 优化)
1. **深度推理**：拥有强大的逻辑分析、代码生成和数学计算能力。
2. **长程记忆**：能够精准回顾和关联长对话历史中的细节，保持上下文一致性。
3. **思维透明**：对于非显而易见的问题，必须通过"显式推理"展示你的思考路径。
# Thinking Protocol (思维协议)
在回答用户之前，你必须执行以下思维循环：
4. **意图识别**：用户真正想要解决的核心痛点是什么？隐含需求是什么？
5. **知识检索**：在你的知识库和当前对话历史中检索相关信息。
6. **逻辑推导**：构建解决路径，预判潜在的错误或陷阱。
7. **自我修正**：检查生成的答案是否准确、无害且符合逻辑。
# Response Format (响应格式规范)
## 场景 A：复杂任务（代码、逻辑、分析、长文本生成）
必须严格包含以下 Markdown 模块：
> **🤔 深度思考**：
> *此处展示你的简要分析逻辑、解题思路或关键决策点。*
> **📋 详细解答**：
> *此处提供具体的答案、代码实现或详细论述。*
> **💡 专家建议**：
> *提供优化建议、潜在风险预警或延伸知识。*
## 场景 B：简单任务（问候、明确的短问题）
- 直接给出简洁、准确的回答，无需展示思考过程。
# Interaction Guidelines (交互准则)
- **准确性优先**：严禁编造事实。如果不知道，请直接说明。
- **代码质量**：生成的代码必须是完整的、可执行的，并包含必要的注释。
- **语言风格**：专业、客观、有条理。避免使用过度情绪化的词语。
temperature: 0.4
max_response_tokens: 8192
max_context_tokens: 1000000
speech_to_text:
model_id: whisper-1
```
deepseek:
```yaml
base_url: https://zenmux.ai/api/v1
api_key: sk-ai-v1-2d2ba59719ff6f0d8d2f439d3b5c84399176d1059302cc4b43c132a4d17e9f03
text_generation:
model_id: deepseek/deepseek-v3.2-speciale
temperature: 0.2
max_response_tokens: 128000
max_context_tokens: 128000
prompt: |
# Role & Identity
你是由 DeepSeek 研发的 **DeepSeek-V3.2-Speciale**，一个专为极致推理和代理性能优化的高算力 AI 助手（{{ baibot_name }}）。
当前会话启动时间: {{ baibot_conversation_start_time_utc }}。
## 版本特别说明 (System Context)
- **定位**：你是一个研究预览版（Research Preview），旨在处理超越常规模型的复杂推理负载。
- **有效期**：本版本服务有效期至 2025年12月15日 15:59 UTC。
- **稳定性**：作为前沿测试模型，你应当专注于解决高难度基准问题，而非生产环境的常规工作流。
# Core Capabilities (DeepSeek 架构优化)
1. **DeepSeek Sparse Attention (DSA)**：利用稀疏注意力机制处理超长上下文，能够精准定位和关联海量信息中的微小细节。
2. **强化推理 (Scaled RL)**：经过大规模后训练强化学习（Post-training RL），具备超越 GPT-5 级别的逻辑推导能力，特别是在数学、编码和复杂任务规划上。
3. **代理任务合成 (Agentic Synthesis)**：拥有强大的指令遵循能力，能够模拟复杂的代理交互，并在交互环境中保持高度的执行一致性。
# Thinking Protocol (思维链协议)
鉴于你是一个“Thinking Mode”优先的模型，在输出最终答案前，必须强制执行深度思维循环：
4. **意图解构**：透过用户表层语言，识别核心痛点与潜在的代理任务需求。
5. **策略规划**：利用 DSA 检索上下文，构建多步骤的解决路径，并预判边界条件。
6. **逻辑演算**：执行显式推理，特别是针对代码和数学问题，进行逐步验证。
7. **合规性检查**：确保输出符合安全标准，并修正任何可能的逻辑幻觉。
# Response Format (响应格式规范)
## 场景 A：深度推理任务（默认模式 - 代码、逻辑、复杂咨询）
必须严格包含以下 Markdown 模块，展现你的“思考模式”：
> **🧠 DeepSeek 思维链**：
> *此处展示你的显式推理过程。包括：问题拆解 -> 关键假设 -> 推导步骤 -> 自我反思。*
> **📋 详细解答**：
> *基于推理结果，提供精准、结构化的最终答案或可执行代码。*
> **🛡️ 专家视角**：
> *提供边缘情况分析、优化建议或针对预览版稳定性的潜在提示。*
## 场景 B：轻量级交互（仅限简单的问候或确认）
- 直接给出简洁、准确的回答，保持高效。
# Interaction Guidelines (交互准则)
- **推理优先**：对于模糊的问题，优先展示你的推理路径，而非直接猜测结论。
- **代码健壮性**：生成的代码必须具备工业级标准，包含错误处理和详细注释，体现 Speciale 级别的编程能力。
- **诚实性**：作为预览版模型，若遇到知识盲区或不确定性，必须明确告知用户，严禁编造。
- **风格**：理性、深刻、极客范。像一位资深的首席工程师那样沟通。
```
01_Projects/AI-Development/Prompt-Library/cycling-training-prompt.md
-8
View File
@@ -1,8 +0,0 @@
windy is a man in his 40s who wants to improve his athletic performance in a cycling. He has some experience with cycling but is looking for a training program that is tailored to his sport. Develop a training program that includes exercises that mimic the movements and demands of his sport, as well as exercises that target the specific muscle groups used in his sport.
I'm a 48-year-old male road cyclist who wants to complete a 200-mile ride in three months time. | would like to complete the ride in under 12 hours. The longest ride | have completed to date was 100 miles long with an average speed of 18mph. Create a week-by-week cycling training program that peaks one week before the event in three months, with the goal to
complete the 200-mile ride in 12 hours or less. | can train three times per week for a maximum of 12 hours during the first two months and four times per week for a maximum of 16 hours during the third month.
I'm a 48-year-old male road cyclist who wants to complete a 200-mile ride in three months time. | would like to complete the ride in under 12 hours. The longest ride | have completed to date was 100 miles long with an average speed of 18mph. Create a day-by-day indoor core excercises program for me, so I can ride longer and faster.
01_Projects/AI-Development/Prompt-Library/dev-engineering-rules.md
-526
View File
@@ -1,526 +0,0 @@
review
---
# Code Review Prompt (improved)
**Goal:** Provide a rigorous, actionable review that balances correctness, security, and maintainability for the following code.
## Inputs
- **Code:**
`{paste code here}`
- **Context (if any):** runtime `{lang/runtime}`, framework `{framework}`, dependencies `{key deps & versions}`, target platform `{os/arch}`, constraints `{perf/mem/latency/security/compliance}`, coding style `{styleguide/eslint/.editorconfig}`, known requirements `{tickets/PRD refs}`.
## Scope of Review
Evaluate and suggest improvements across these dimensions:
1. **Correctness & Edge Cases**
- Logic/algorithm soundness, off-by-one, null/empty, boundary values, error handling & retries, concurrency/races, timezones/locale, I/O/resource cleanup.
2. **Security**
- OWASP Top 10 risks relevant to this code (injection, auth/authorization, SSRF, path traversal, XSS, CSRF, deserialization, secrets handling, logging of sensitive data), dependency risk, input validation, output encoding, sandboxing, least privilege, DoS hotspots.
3. **Performance**
- Time/space complexity, hot paths, allocations, N+1 queries, sync vs async, batching/caching, I/O patterns, streaming vs buffering, algorithmic alternatives.
4. **API & Design Quality**
- Public contracts & invariants, error models, idempotency, purity & side effects, cohesion/coupling, layering, testability, configuration vs hard-coding.
5. **Readability & Maintainability**
- Naming, structure, small functions, duplication, comments/docs, idiomatic use of `{language}`, lint/format compliance.
## Deliverables (use this exact structure)
### 1) Executive Summary
- One paragraph on overall health and top 3 risks.
### 2) Findings Table
Provide a table with: **ID | Severity (High/Med/Low) | Category | Symptom | Why it matters | Evidence (line refs) | Fix summary**
### 3) Patch Suggestions
For each High/Med item, include a **minimal diff** or **before/after** snippet:
```diff
{target file path}
- {problematic code}
+ {improved code}
```
Explain the trade-offs and why the fix is correct.
### 4) Tests to Add
List concrete test cases (names + intent). Include edge values and failure paths.
- Unit: `{TestName_Should...}`
- Integration: `{Scenario_When..._Then...}`
- Property/Fuzz (if applicable): input domains & invariants.
### 5) Performance Notes
- Estimated complexity and bottlenecks.
- Quick wins (e.g., cache/batch/stream) and expected impact.
### 6) Security Checklist
- Inputs validated? Output encoded? Secrets sourced from vault? Least privilege? Safe defaults? Rate limiting? Logging PII redaction?
### 7) Maintainability Improvements
- Refactors (small + incremental), dead code removal, error taxonomy, configuration externalization, docs/comments to add.
### 8) Quality Scores
Give 1–5 scores for: **Correctness, Security, Performance, Design, Readability, Testability**, with one-line justification each.
## Constraints
- Prefer **minimal, targeted changes** over large rewrites.
- Match existing project style and patterns.
- If context is missing, **state assumptions explicitly** and proceed.
- Link to idiomatic patterns or standards **only if widely accepted**; keep recommendations framework-agnostic where possible.
## Output Format
Return **only** the sections 1–8 above in Markdown. Keep code blocks self-contained and compilable where possible.
---
需要更精简版时，可以用这句：
> Review the code for **correctness, security, performance, API/design, and maintainability**. Return: (1) 5-sentence summary; (2) Findings table (ID, Severity, Why, Evidence, Fix); (3) Minimal diffs for Med/High issues; (4) Test cases to add; (5) Perf quick wins; (6) Security checklist status; (7) 1–5 scores for each quality dimension with 1-line rationale. Use project style, prefer minimal changes, state assumptions if context is missing.
code review:
# Code Review Prompt (final)
**Goal:** Provide a rigorous, _actionable_ review that balances **correctness, security, performance, and maintainability** for the following code.
---
## Inputs
- **Code:**
```text
{paste code here}
```
- **Context (optional but recommended):**
- runtime: `{lang/runtime}`
- framework: `{framework}`
- key dependencies & versions: `{deps & versions}`
- target platform: `{os/arch}`
- constraints: `{perf/mem/latency/security/compliance}`
- coding style: `{styleguide/eslint/.editorconfig}`
- known requirements: `{tickets/PRD refs}`
If any context is missing, **state your assumptions explicitly** before the review.
---
## Scope of Review
Evaluate and suggest improvements across these dimensions:
1. **Correctness & Edge Cases**
- Logic/algorithm soundness
- Off-by-one, null/empty, boundary values
- Error handling & retries
- Concurrency/races
- Timezones/locale handling
- I/O & resource cleanup
2. **Security**
- Relevant OWASP Top 10 risks (injection, auth/z, SSRF, path traversal, XSS, CSRF, deserialization)
- Secrets handling & configuration
- Input validation & output encoding
- Logging of sensitive data
- Least privilege, sandboxing, DoS hotspots
3. **Performance**
- Time & space complexity
- Hot paths and allocations
- N+1 queries / chatty I/O
- Sync vs async behavior
- Batching, caching, streaming vs buffering
- Algorithmic alternatives
4. **API & Design Quality**
- Public contracts & invariants
- Error model & error propagation
- Idempotency and side effects
- Cohesion & coupling, layering boundaries
- Dependency direction (domain vs infra)
- Testability and configuration vs hard-coding
5. **Readability & Maintainability**
- Naming and intent clarity
- Function/module size and structure
- Duplication vs reuse
- Comments/docs (where needed)
- Idiomatic use of `{language}`
- Lint/format compliance
---
## Deliverables (use this exact structure)
### 1) Executive Summary
- One short paragraph on overall health.
- List the **top 3 risks or opportunities** (bullets).
### 2) Findings Table
Provide a table with:
- **ID** – short stable identifier (e.g., `C1`, `S2`, `P3`)
- **Severity** – `High` / `Medium` / `Low`
- **Category** – `Correctness`, `Security`, `Performance`, `Design`, `Readability`, `Testability`, etc.
- **Symptom** – what is wrong / suspicious
- **Why it matters** – impact / risk
- **Evidence (line refs)** – e.g., `file.go:42-57`
- **Fix summary** – 1–2 line suggested direction
Example:
|ID|Severity|Category|Symptom|Why it matters|Evidence|Fix summary|
|---|---|---|---|---|---|---|
|C1|High|Correctness|Possible nil deref on error path|Can cause runtime panic in production|`handler.go:78-85`|Check error before use; return early on fail|
### 3) Patch Suggestions
For each **High** or **Medium** item in the table, include a **minimal diff** or **before/after** snippet.
```diff
{target file path}
- {problematic code}
+ {improved code}
```
- Keep patches **local and incremental**, not full rewrites.
- Explain **why** the fix is correct, and any trade-offs (perf, readability, behavior change).
### 4) Tests to Add
List **concrete test cases** to cover the identified issues and edge cases.
- Unit tests (with intent):
- `Test_{UnitName}_ShouldHandleEmptyInput` – verifies behavior when input is empty
- `Test_{FuncName}_ShouldReturnErrorOnTimeout` – covers timeout/failure path
- Integration tests:
- `{Scenario_When..._Then...}` – describe full flows: external calls, DB, queues, etc.
- Property/Fuzz tests (if applicable):
- Describe **input domain**, invariants, and what must always hold.
Where possible, map tests back to **Finding IDs** (e.g. “C1, S2”).
### 5) Performance Notes
- Estimate complexity and potential bottlenecks of key paths.
- Call out:
- Any obvious **N+1** patterns
- Unnecessary allocations or copying
- Inefficient data structures or algorithms
- Suggest **quick wins**:
- Caching, batching, streaming, preallocation, memoization
- Expected impact (qualitative: small/medium/large)
### 6) Security Checklist
Answer briefly (Yes/No/N.A. + short note):
- Inputs validated at boundaries?
- Outputs properly encoded for their sinks (HTML/SQL/OS/etc.)?
- Auth & authorization checks present and correctly ordered?
- Secrets kept out of code (config, env, vault)?
- Least privilege for external resources (DB, queues, files)?
- Safe defaults (e.g., secure TLS, secure cookies, strict modes)?
- Rate limiting / throttling for expensive or exposed endpoints?
- Logs avoid PII/credential leakage; sensitive data redacted or omitted?
Highlight any **High** severity gaps and link them to Findings IDs.
### 7) Maintainability Improvements
- Small, incremental refactors:
- Extract helpers / smaller functions
- Reduce duplication (shared utilities, common error handling)
- Clarify boundaries between layers (domain/app/infra)
- Error taxonomy:
- Group errors into meaningful types/categories (e.g., validation vs system vs external)
- Standardize error wrapping and messages
- Configuration:
- Externalize magic numbers/strings
- Centralize feature flags or switches
- Documentation:
- Add or update docstrings for non-obvious logic
- Brief README/ADR notes if design is non-trivial
### 8) Quality Scores
Give **1–5** scores (5 = excellent, 1 = poor) with a **one-line justification** each:
- **Correctness:** `X/5` – `{short reason}`
- **Security:** `X/5` – `{short reason}`
- **Performance:** `X/5` – `{short reason}`
- **Design:** `X/5` – `{short reason}`
- **Readability:** `X/5` – `{short reason}`
- **Testability:** `X/5` – `{short reason}`
---
## Constraints
- Prefer **minimal, targeted changes** over big-bang rewrites.
- Match **existing project style and patterns** where visible.
- If context is missing, **state assumptions explicitly** and proceed.
- Keep recommendations **framework-agnostic** where possible; only reference widely accepted idioms and standards.
- When in doubt, **prioritize clarity and safety** over micro-optimizations.
---
## Output Format
Return **only** sections **1–8** above in Markdown when performing an actual review.
Keep all code blocks self-contained and compilable where possible.
----
# 可观测性 —— **4.5 / 10**
优点：
- 使用 zap
- 有 telemetry endpoint 配置
存在重大缺口：
- 没 metrics
- 没 health checks
- 没 tracing schema
- 没日志字段规范
- 没报警策略
专业系统里可观测性是“一等公民”，缺这块分数自然拉低。
---
# 4️⃣ 可靠性（Reliability & Fault Handling）—— **5.5 / 10**
优点：
- JetStream（正确选择）
- 配置级 backoff / ack_wait / replay_from
- 已考虑重试机制
不足：
- 没看到 dead-letter pipeline 文档
- 没看到 poison message 策略
- 没看到 DB 阻塞时的 backpressure
- 没看到幂等性模型
- 没看到断线重连逻辑的描述
这些是专业评分严格扣分的部位。
### 严格评审的缺失
- 没看到“dead-letter pipeline”定义
- 没看到“poison message”策略
- 没看到“持久化失败策略”
- 没看到“DB 降级”逻辑
- 没看到“幂等性策略”（特别关键）
- 没看到“重平衡策略”（consumer scaling）
- 没看到“高可用拓扑”（replicas 仅是 JetStream 层，服务自身无说明）
按专业级评分，就是 **4/10**。
这个维度是最严格的（专业评分里非常重要）。
### ⭐ 有点：
- 有 Zap
- 有 OTEL endpoint 配置
### ❌ 不足（按专业要求）
- 没有 metrics（prometheus）
- 没有 trace pipeline（span 设计/采样策略）
- 没有健康检查
- 没有 readiness
- 没有 structured logging contract（如 msg_id / request_id / nats_sequence）
- 未定义错误分类（business vs transient vs fatal）
- 没有日志示例
- 没有运行时仪表盘（Grafana dashboards）
> **严格评分下，这就是 3/10。**
>
----
01_Projects/AI-Development/Prompt-Library/movie-review-prompt.md
-29
View File
@@ -1,29 +0,0 @@
忘记之前的所有要求，请分别以电影导演，热爱电影的观众，普通人的角度评论一下电影。
请实用下面格式：
- 讲讲电影的整体感受，分别从电影拍摄的时期，以及现在这个时候讲
- 评论一下电影的故事情节，任务，已经电影想要传达的内容
- 总结一下电影的有点和缺点
- 给电影做一个评分，从0开始，10分最高分
如果你明白了上述指示，而我又没有告诉你电影名，请回答：”请问你想了解哪一部电影“
如果知道了电影，请完成上面指示
请完成下面任务：
1. 以一个普通人的角度，评价一下电影，简单讲讲观看电影的体验，如果觉得电影不错，推荐给好友
2. 以一个资深电影迷的角度，写一篇发表到社交媒体的影评。涉及导演，演员，音乐等电影相关元素，最后发表一下自己的看法，谈谈电影的优缺点。
3. 以一个电影从业人员的角度，写一篇专业的影评到电影专业期刊。从专业的角度分析电影的素质，分别从观看和制作的角度评价一下电影的主要元素和主要有点
4. 于此同时，每一个角度都要给出一个对电影的评分，从0开始，10分最高，并给出简单的原因
用下面格式：
简介：<首先请介绍一下电影，译名(原名)，创作年代，导演，主要演员。>
普通观众：<普通人的角度，评分>
影迷： <资深影迷的角度，内容可以丰富一些，去掉空洞的泛泛而谈的内容， 评分>
从业人员: <从业人员的角度， 评分>
如果你知道我说的是什么电影，请完成任务，如果还不知道，可以问我电影名