SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsday2-create-context-sync-skill

day2-create-context-sync-skill

General & other

View repository

Updated 10-15 18:31

About this skill

Day 2: Creating Your Own Context Sync Skill

When this skill is invoked, you must follow the STOP PROTOCOLbelow.

Terminology

Key terms used in this skill:

Term Description
MCP A channel through which Claude converses with external services (Slack, Gmail, etc.). Extending the "tools" learned in Day 1 to the outside
subagent Claude calling another Claude to do work. Used when handling multiple tasks simultaneously
Explore agent A specialized subagent that figures out the project folder structure
API A data gateway provided by a service. A way to fetch data directly with code when there is no MCP
Skill A document that teaches Claude Code how to perform a specific task. The one experienced in Day 1 Block 3-2

STOP PROTOCOL — Never Violate

This protocol is the top-priority rule of this skill. Violating the rules below will ruin the class.

Each block must proceed over exactly 2 turns

┌─ Phase A (첫 번째 턴) ──────────────────────────────┐
│ 1. references/에서 해당 블록 파일의 EXPLAIN 섹션을 읽는다    │
│ 2. 기능을 설명한다                                        │
│ 3. references/에서 해당 블록 파일의 EXECUTE 섹션을 읽는다    │
│ 4. "지금 직접 실행해보세요"라고 안내한다                     │
│ 5. ⛔ 여기서 반드시 STOP. 턴을 종료한다.                    │
│                                                          │
│ ❌ 절대 하지 않는 것: 퀴즈 출제, QUIZ 섹션 읽기             │
│ ❌ 절대 하지 않는 것: AskUserQuestion 호출 (Block 0,2,4 제외)│
│ ❌ 절대 하지 않는 것: "실행해봤나요?" 질문                   │
└──────────────────────────────────────────────────────────┘

  ⬇️ 사용자가 돌아와서 "했어", "완료", "다음" 등을 입력한다

┌─ Phase B (두 번째 턴) ──────────────────────────────┐
│ 1. references/에서 해당 블록 파일의 QUIZ 섹션을 읽는다       │
│ 2. AskUserQuestion으로 퀴즈를 출제한다                     │
│ 3. 정답/오답 피드백을 준다                                 │
│ 4. 다음 블록으로 이동할지 AskUserQuestion으로 묻는다         │
│ 5. ⛔ 다음 블록을 시작하면 다시 Phase A부터.                │
└──────────────────────────────────────────────────────────┘

Key Prohibitions (Never Violate)

  1. Do not call AskUserQuestion in Phase A (except Blocks 0, 2, 4) — these 3 blocks are exceptions because user selection is required
  2. Do not give a quiz in Phase A — read the QUIZ section only in Phase B
  3. Do not ask "Did you try running it?" in Phase A — wait until the user speaks first
  4. Do not do EXPLAIN + QUIZ at the same time in one turn — always split into 2 turns

Output official documentation URLs (never omit)

At the start of Phase A of every block, output the > 공식 문서: URL을 at the top of the corresponding reference file exactly as is.

📖 공식 문서: [URL]
  • If there are multiple URLs in the reference file, output all of them
  • Do not summarize or omit URLs

Required phrase at the end of Phase A

At the end of Phase A, you must output a phrase in the form below and Stop:

---
👆 위 내용을 직접 실행해보세요.
실행이 끝나면 "완료" 또는 "다음"이라고 입력해주세요.

After this phrase, do not output any tool calls (including AskUserQuestion) or additional text.

Core Strategy: Template First, Step-by-Step Customization

Proceed in the following way:

  1. When the user selects a tool in Block 0, templates/context-sync.md immediately create the skill file based on
  2. Afterward, in Blocks 1–5, modify/expand only the relevant parts of the created skill
  3. Finally, run the completed skill and check the result

The template includes examples for 4 tools: Slack, Notion, Gmail, and Google Calendar. Keep only the parts needed for the tool combination the user selected, and add new tools using the same pattern.

Block-Template Section Mapping

Template areas modified in each block:

Block Modification Target Template Section
0 Generate skill skeleton Entire (keep only selected tools)
1 Reflect project context frontmatter description, collection scope
2 Finalize connection method "Collection Method" for each source
3 Execute & verify collection "Execution Flow" section + adjust "Information to Extract"
4 Set output format "Output Format" section
5 Final cleanup + execution Overall wrap-up

Block-Specific Rules

  • Block 0 (Tool selection + skill generation): In Phase A, explain + select tools with AskUserQuestion. Generate a template-based skill from the selection results → Stop. In Phase B, a quiz on the generated skill.
  • Block 1 (Project exploration): In Phase A, understand the project structure with the Explore agent and share the results → Stop. In Phase B, a quiz.
  • Block 2 (Tool connection): In Phase A, guidance on choosing MCP vs API + AskUserQuestion → Claude performs the setup on the user's behalfand the user checks the results → Stop. In Phase B, a quiz.
  • Block 3 (Execute & verify collection): In Phase A, explain parallel collection with subagents + execute → show collection results classified as success/failure → retry failed sources + check collected data quality → Stop. In Phase B, a quiz.
  • Block 4 (Output settings): In Phase A, guidance on selecting Output format + AskUserQuestion → modify the skill according to the selection → Stop. In Phase B, a quiz.
  • Block 5 (Completion + execution): In Phase A, summarize the final skill configuration + actual execution → Stop. In Phase B, a comprehensive quiz + wrap-up.

Block 0 Exception Rules

Block 0의 Phase A는 Use AskUserQuestion. Since the tool selection is a prerequisite for all subsequent blocks, user input must be obtained.

Phase A progression order:

  1. references/block0-tool-selection.mdRead and explain the EXPLAIN section of
  2. Receive the tool selection via AskUserQuestion (multiSelect: true)
  3. templates/context-sync.md Read the template
  4. In the user's project, according to the selected tools, .claude/skills/my-context-sync/SKILL.mdcreate
  5. Briefly show only the overall structure of the generated file and Stop (details are covered in later blocks)

Block 2 Exception Rules

Block 2의 Phase A도 Use AskUserQuestion. For each tool, the connection method must be selected between MCP and API.

Core principle: Claude performs the setup on the user's behalf, and the user checks the results.

When MCP is selected:

  1. references/block2-tool-connection.mdFollow the MCP guidance in
  2. scripts/mcp_servers.pyuse to search for a suitable MCP server in the GitHub README.md
  3. Show the search results, and Claude .mcp.jsonregister the server in
  4. /mcp also check the server connection status with the command

When selecting API:

  1. Claude writes the API call code directly
  2. the user's skill scripts/ save to the folder

Block 4 Exception Rules

Block 4의 Phase A도 use AskUserQuestion. You must select the Output format.

References File Map

Block File Content
Block 0 references/block0-tool-selection.md Tool selection + template-based skill generation
Block 1 references/block1-project-explore.md Identify project structure with the Explore agent
Block 2 references/block2-tool-connection.md Select MCP vs API connection method + execute
Block 3 references/block3-parallel-collection.md subagent parallel collection + result verification
Block 4 references/block4-output-format.md Select Output format (markdown, Slack, Notion)
Block 5 references/block5-finalize.md Finalize skill + execute + wrap up

File paths are relative to this SKILL.md. Each reference file consists of the ## EXPLAIN, ## EXECUTE, ## QUIZ section.

Templates / Scripts File Map

File Purpose
templates/context-sync.md Context Sync skill default template (includes 4 types: Slack, Notion, Gmail, GCal)
scripts/mcp_servers.py Search for MCP servers on GitHub + parse README.md + installation guide

Collection scripts for Gmail/Calendar etc. are written directly by Claude in Block 2 according to the user's choices.

Progress Rules

  • Proceed one block at a time
  • Move with "next", "skip", or block number/name
  • Gradually modify the skill file created in Block 0 in subsequent blocks
  • In the user's project .claude/skills/my-context-sync/ create the skill in the directory
  • Using the Explore agent and subagents is key, so actively utilize them
  • If a question about Claude Code comes up, answer with the claude-code-guide agent (built-in tool). After answering, guide the user step by step so they can follow along themselves, and use AskUserQuestion when asking questions. If you determine that the built-in agent's answer is inaccurate, save the official documentation to a file as curlthen read it carefully with the Read tool and answer again with accurate information

Start

When starting the skill, first install the latest curriculumand then select a block.

Step 1: Install the latest skill

Output the command below and run it with Bash:

npx skills add ai-native-camp/camp-1 --agent claude-code --yes

Briefly inform the user of the execution result (e.g., "4 skills have been installed at the latest version").

Step 2: Select Block

Show the table below and ask with AskUserQuestion where to start.

Block Topic Content
0 Tool selection Choose tools to sync + generate skill skeleton
1 Project exploration Identify project structure with Explore
2 Tool Connection Connect tools via MCP or API
3 Run Collection & Verify Parallel collection via subagents + result verification
4 Output Settings Select output format + modify skill
5 Complete + Run Run the final skill + wrap up 
AskUserQuestion({
  "questions": [{
    "question": "Day 2: 나만의 Context Sync 스킬 만들기\n\n어디서부터 시작할까요?",
    "header": "시작 블록",
    "options": [
      {"label": "처음부터 (Block 0)", "description": "sync할 도구 고르기 + 스킬 골격 생성"},
      {"label": "도구 연결 (Block 2)", "description": "도구 선택은 했고, MCP/API 연결부터"},
      {"label": "수집 실행 & 검증 (Block 3)", "description": "연결 완료, 수집부터"},
      {"label": "Output 설정 (Block 4)", "description": "수집 완료, 출력 형식부터"}
    ],
    "multiSelect": false
  }]
})

After selecting a starting block → proceed from that block's Phase A.