메뉴
Load project context by reading README and recursively following relevant documentation links based on a question or topic
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/nicolasdao/puffy-core --skill init-context명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | init-context |
| description | Load project context by reading README and recursively following relevant documentation links based on a question or topic |
| allowed-tools | Read, Glob, Grep |
| argument-hint | ["question or topic"] |
You are being asked to load context about this project to help answer a specific question or explore a particular topic.
IMPORTANT:
1. You are ONLY gathering information. DO NOT modify any files, make any changes, or suggest changes yet. This is purely a research and context-loading phase.
2. DO NOT READ ANY FILES LOCATED UNDER THE specs/ FOLDER OR UNDER THE docs/manual/ FOLDER
3. The following files MUST be read regardless of the topic or question if they exist under the docs/ folder:
** - gotchas.md or GOTCHAS.md — critical project-specific pitfalls and edge cases (see Gotchas Loading for hub+domain handling)**
** - mission.md or MISSION.md — project vision, values, and decision-making compass**
These files contain foundational context that is always relevant.
The user has provided this question or topic:
$ARGUMENTS
Follow this iterative process to build up context:
Start by reading the README.md file (if it exists in the current directory). This is your entry point to understand:
Gotchas are project-specific pitfalls — things that WILL bite you if ignored. They must be loaded early and treated as high-priority warnings, not background documentation.
Detection: After reading README.md, determine which gotchas format this project uses:
docs/gotchas/ directory using Glob: docs/gotchas/*.mddocs/gotchas.md (or docs/GOTCHAS.md)Format A — Hub + Domain files (the docs/gotchas/ directory exists):
docs/gotchas.md). This is a small index (~30-50 lines) listing all gotcha domains with descriptions and links to docs/gotchas/<domain>.md files. Always read it in full.docs/gotchas/<domain>.md files are relevant to the user's question/topic. Use the same relevance criteria as Step 2 (link evaluation). Read only the relevant domain file(s).Format B — Monolithic file (no docs/gotchas/ directory, only docs/gotchas.md):
Read docs/gotchas.md in full. This is the legacy format — the entire file is loaded regardless of size.
Attention framing: When you encounter gotchas content (from either format), treat each gotcha as a warning. These are not reference documentation — they are lessons learned from production incidents. In your summary, present gotchas prominently under their own heading, not buried inside general findings.
From the README.md, identify all links to other documentation files (markdown files, text files, or other documentation). For each link, evaluate whether it's relevant to the user's question/topic.
Relevance criteria:
Track what you've read:
Read each relevant documentation file you identified. As you read each file:
Continue the process:
Important safeguards:
After completing the recursive documentation discovery, provide:
Gotchas (WARNINGS): If any gotchas were loaded, present them FIRST under a dedicated heading. These are not suggestions — they are hard-won lessons. Frame them as: "The following gotchas are directly relevant to this work and must be respected." List each gotcha with its title and a one-line summary of the risk.
List of files read: Show which documentation files you read and in what order.
Why each was relevant: Brief explanation of why you chose to read each file.
Key findings: Summarize the information relevant to the user's question/topic.
Context loaded: Confirm that you now have sufficient context to help with their question.
Progressive discovery reminder (hub+domain gotchas format only): If the project uses the docs/gotchas/ directory structure and some domain files were NOT loaded, include this notice at the end of the summary:
Additional gotcha domains available: [list the unloaded domain names with one-line descriptions from the hub]. If the conversation shifts to any of these areas, load the relevant
docs/gotchas/<domain>.mdfile before proceeding — these contain critical warnings that could prevent mistakes.
Omit this notice if all domain files were loaded, or if the project uses the monolithic gotchas format (Format B).
DO NOT make any changes or suggestions yet. Simply state that you've loaded the context and are ready to help with the user's question or task.
User question: "How does the deployment pipeline work?"
1. Read README.md
→ Find link to docs/architecture.md (relevant ✓)
→ Find link to docs/api.md (not relevant ✗)
→ Find link to docs/deployment.md (relevant ✓)
2. Gotchas Loading
→ Glob docs/gotchas/*.md → found: database.md, deployment.md, frontend.md
→ Hub+domain format detected
→ Read docs/gotchas.md (hub — always read)
→ Read docs/gotchas/deployment.md (relevant to question ✓)
→ Skip: database.md, frontend.md (not relevant to question)
→ Record unloaded domains: database, frontend
3. Read docs/architecture.md
→ Find link to docs/data-model.md (not relevant ✗)
4. Read docs/deployment.md
→ No new relevant links found
5. Done — summarize findings
→ Present deployment gotchas as WARNINGS first
→ List files read and key findings
→ Include progressive discovery reminder:
"Additional gotcha domains available: database, frontend.
If work shifts to these areas, load the relevant gotcha file first."
Remember: This is a context-loading phase only. No modifications, no suggestions, just information gathering.