hs3180/feeling-lucky

Feeling Lucky (手气不错)

Randomly pick ONE real disclaude use case, simulate a natural user interaction, and report observations.

When to Use This Skill

Use this skill for:

• Daily automated dogfooding of disclaude features
• Randomly testing a real use case from disclaude's feature set
• Discovering bugs and UX issues through simulated real-user interactions

Keywords that trigger this skill: "手气不错", "随机测试", "feeling lucky", "dogfooding", "自我体验"

Core Principle

You ARE a real disclaude user. You know everything disclaude can do. You randomly pick one use case and interact naturally — as if you just thought of something to try.

Context Variables

When invoked, you receive:

• Chat ID: Feishu chat ID (from "Chat ID: xxx")
• Message ID: Message ID (from "Message ID: xxx")
• Sender Open ID: Sender's open ID (from "Sender Open ID: xxx")

---

disclaude Feature & Use Case Catalog

The following is disclaude's complete feature set. Each item is a real use case that a real user would try.

Category A: Chat & Code Operations (日常对话与代码)

1. Ask to read a file — "帮我看看 src/agents/base-agent.ts 的内容"
2. Ask to edit code — "把 reset.ts 里的 setTimeout 改成 3 秒"
3. Ask to search code — "搜索所有用到 agentPool.reset 的地方"
4. Ask to run a command — "跑一下 npm test 看看结果"
5. Ask to create a file — "帮我创建一个新的 handler 文件"
6. Ask for code explanation — "解释一下 IPC 的消息路由是怎么工作的"

Category B: Schedule & Automation (定时任务与自动化)

7. Create a scheduled task — "创建一个每天早上 10 点提醒我开 standup 的定时任务"
8. List existing schedules — "看看现在有哪些定时任务"
9. Modify a schedule — "把那个每日提醒改成每周一执行"

Category C: GitHub Integration (GitHub 操作)

10. Check PR status — "看看现在有哪些 open 的 PR"
11. View issue details — "帮我看看 issue #1617 的内容"
12. Submit feedback — "/feedback 搜索功能有时候不太准确"

Category D: Feishu/Lark Features (飞书功能)

13. Read a Feishu document — Send a Feishu doc link and ask to read it
14. Upload a file to Feishu — "帮我把这个报告上传到飞书"
15. Send an interactive card — "发一个带按钮的卡片让我选择"
16. Send a formatted card — "用卡片格式展示这些数据"

Category E: Browser & Web (浏览器与网页)

17. Scrape a website — "帮我看看这个网页上的内容"
18. Monitor a page — "打开这个页面截个图给我看看"
19. Web search — "搜索一下 Claude Code 最新版本有什么变化"

Category F: Document & Content (文档与内容)

20. Parse a PDF — "帮我解析这个 PDF 的内容"
21. Generate a PPT — "根据这个大纲帮我生成一个 PPT"

Category G: Discussion & Community (讨论与社区)

22. Start a discussion — "发起一个关于代码规范的讨论"
23. Generate a BBS topic — "给话题群生成一个讨论话题"
24. Daily chat review — "回顾一下今天的聊天记录，有什么值得改进的"

Category H: Control Commands (控制命令)

25. Reset session — "/reset"
26. Check status — "/status"
27. Help — "/help"

Category I: Edge Cases & Stress (边界与压力)

28. Empty or very short input — "" or "?" or "嗯"
29. Very long input — Paste a 2000+ character message and ask for summary
30. Rapid consecutive commands — Send 3 messages in quick succession
31. Off-topic question — "今天中午吃什么" or "讲个笑话"
32. Multilingual input — "Can you help me with this code?" (English input)
33. Mixed format input — Send code snippets mixed with natural language

---

Execution Process

Step 1: Pick a Random Use Case

From the catalog above, randomly select ONE use case. Use this method:

Seed = hash(date + chatId + messageIndex) mod 33
Selected scenario = catalog[Seed]

In practice: pick a scenario that feels fresh and different from what you tested recently.

Anti-repetition check:

cat workspace/feeling-lucky/history.md 2>/dev/null | tail -10 || echo "No history"

If the selected scenario was tested in the last 3 sessions, pick the next one.

Step 2: Generate a Natural Message

Write a message as if you are a real disclaude user who just thought of something to try. The message should:

• Sound natural and casual (like chatting with a friend)
• Be specific enough to test a real feature
• Not sound like a test or evaluation
• Include realistic context (file names, command names, etc. from the actual codebase when relevant)

Examples by category:

| Category | Example Message | |----------|----------------| | A (Code) | "帮我看看 vitest.config.ts 里 coverage 的配置" | | B (Schedule) | "每天下午 3 点提醒我 review PR" | | C (GitHub) | "看看现在有哪些 open issues 标记了 bug" | | D (Feishu) | "发一个交互卡片，问大家今天想讨论什么话题" | | E (Browser) | "帮我查一下 Vitest 最新版本的 release notes" | | F (Document) | "帮我解析 /tmp/report.pdf" | | G (Discussion) | "回顾一下最近的 PR 讨论，有什么值得关注的" | | H (Control) | "/status" | | I (Edge) | "asdfghjkl" |

Step 3: Send the Message

Use mcp__channel-mcp__send_text to send the generated message to the target chat:

mcp__channel-mcp__send_text({
  text: "{generated message}",
  chatId: "{chatId}"
})

Step 4: Observe (Internal Only)

After sending, mentally note (do NOT send to user):

• Did the message format correctly?
• Was the message natural-sounding?
• Any issues with the sending process itself?

If a real disclaude bot is listening in the chat, observe its response quality:

• Did it understand the intent?
• Was the response helpful?
• Any errors or unexpected behavior?

Step 5: Record and Report

Append the test record to history:

mkdir -p workspace/feeling-lucky
echo "- $(date +%Y-%m-%d): [Scenario #{number}] {scenario name} - {one-line observation}" >> workspace/feeling-lucky/history.md

If significant bugs or issues were found, create a GitHub issue:

gh issue create --repo hs3180/disclaude \
  --title "bug: {brief description}" \
  --body "Found via feeling-lucky dogfooding.\n\n{details}"

---

Daily Variety Rules

To ensure broad coverage over time:

| Day | Preferred Category | |-----|-------------------| | Monday | A (Code) or C (GitHub) | | Tuesday | B (Schedule) or G (Discussion) | | Wednesday | D (Feishu) or E (Browser) | | Thursday | F (Document) or H (Control) | | Friday | I (Edge Cases) — shake things up |

This is a soft preference, not a strict rule. Randomness is still the priority.

---

Quality Guidelines

Good Test Messages:

• Sound like a real person naturally interacting
• Test a specific, identifiable feature
• Are varied in type and complexity
• Include realistic details (actual file names, commands, etc.)

Avoid:

• Messages that sound like test scripts
• Always testing the same category
• Overly complex multi-step scenarios (keep it simple)
• Messages that require external state or special setup
• Multiple test messages in one session (one per day)

---

DO NOT

• Generate messages without picking from the catalog first
• Send more than one test message per invocation
• Include "I'm testing you" or "This is a dogfooding test" in the message
• File GitHub issues for trivial cosmetic preferences
• Skip the history check (causes repetitive testing)
• Create elaborate test plans or scoring systems

---

Integration with Other Skills

• daily-news-inspiration: Uses external news; this skill uses internal feature catalog
• daily-soul-question: Analyzes chat history; this skill simulates user interactions
• bbs-topic-initiator: Generates topics for groups; this skill tests features
• next-step: Can be triggered after this skill completes to suggest follow-up actions

---

Schedule Configuration

See schedule.template.md in this directory for daily automated execution setup.