dev-project-builder
Create Claude-Project-like experiences inside Claude Code — persistent context, specialized responses, and working memory for a recurring workspace. Use when the user says "make a project", "persisten
SKILL.mdskills/dev-project-builder/SKILL.md
GitHubProject Builder
A methodology for creating "Claude Project-like" skills that provide persistent context, specialized behavior, and true memory through vault integration.
The Pattern
Claude Projects provide: uploaded files, custom instructions, persistent memory, and specialized responses. In Claude Code, we replicate this with a 5-component structure:
┌─────────────────────────────────────────────────────────────┐
│ PROJECT STRUCTURE │
├─────────────────────────────────────────────────────────────┤
│ 1. Memory Note (the "persistent context") │
│ └── Background, current state, breakthroughs, patterns │
│ └── Read on EVERY activation, update when insights emerge│
│ └── Location: 20_Areas/[area]/[name]-memory.md │
│ │
│ 2. Core Document (the "map") │
│ └── Goals, progress, key reference info │
│ └── Location: 20_Areas/ or 10_Projects/ │
│ │
│ 3. Reference Files (the "uploads") │
│ └── Frameworks, theory, templates, guides │
│ └── Location: 30_Resources/ or 20_Areas/ │
│ │
│ 4. Working Folder (the "session history") │
│ └── Logs, session notes, drafts │
│ └── Location: 20_Areas/ or 50_Journal/ │
│ │
│ 5. Skill Definition (the "instructions") │
│ └── Modes, templates, triggers, style guide │
│ └── Location: .claude/skills/NAME/SKILL.md │
└─────────────────────────────────────────────────────────────┘
Memory Note vs Core Document
| Aspect | Memory Note | Core Document |
|---|---|---|
| Purpose | Persistent context about the user | Reference info and plans |
| Updates | Frequently (after breakthroughs) | Less frequently |
| Content | Background, state, patterns, learnings | Goals, frameworks, timelines |
| Analogy | Claude Project's Memory feature | Uploaded reference files |
When to Use This Skill
- User says "I want to create a skill for X"
- User says "帮我做一个像 Claude Project 那样的..."
- User has a complex domain with existing vault content that needs specialized AI interaction
Workflow: Creating a New Project-Like Skill
Step 1: Identify the Domain
Ask the user:
"What topic or area do you want Claude to help with? What kind of companion or assistant are you looking for?"
Step 2: Find Existing Content
Search the vault for related content:
- Look in
20_Areas/for existing area documentation - Look in
30_Resources/for reference materials - Look in
50_Journal/for existing logs or entries - Look in
10_Projects/for active project plans
Report what you find:
"I found these related files in your vault: [list]. Which of these are relevant to your new skill?"
Step 3: Map to Structure
Help the user identify each component:
| Component | Question to Ask | Example |
|---|---|---|
| Memory Note | "Where should I store persistent context about you?" | healing-memory.md |
| Core Document | "What's the main 'map' that tracks your state/progress?" | 焦虑型依恋-我的疗愈地图.md |
| Reference Files | "What frameworks, guides, or theory should I reference?" | CBT prompts, methodology docs |
| Working Folder | "Where should I write logs and session notes?" | 疗愈日志/, sessions/ |
| Modes | "What different ways do you want to interact?" | Trigger processing, check-in, review |
Memory Note Template
---
type: [domain]-memory
updated: YYYY-MM-DD
---
# [Domain] Memory
## Purpose & Context
[Why this matters, current life situation, background]
## Current State
[What's being actively worked on, recent focus areas]
## Core Patterns & Insights
[Identified patterns, key beliefs, recurring themes]
## Key Breakthroughs
[Significant insights with dates]
- YYYY-MM-DD: [breakthrough description]
## Approach & What Works
[Techniques, methods, phrases that help]
## Tools & Resources
[Books, frameworks, preferred approaches]
Step 4: Define Modes
For each mode, define:
- Trigger phrases: What the user says to activate this mode
- Workflow: Step-by-step process
- Output: What gets created (logs, updates, etc.)
Common mode patterns:
- Processing mode: Work through something (e.g., trigger, problem, idea)
- Check-in mode: Quick status/energy check
- Review mode: Analyze patterns over time
- Crisis mode: Urgent support with grounding
- Reference mode: Look up and apply frameworks
Step 5: Define Style Guide
Ask:
"How should Claude respond in this domain? What tone, pacing, or approach works best?"
Common style elements:
- Communication tone (warm, direct, professional, etc.)
- Response pacing (one question at a time vs. comprehensive)
- Language preference (Chinese, English, bilingual)
- Key phrases or affirmations to use
Step 6: Generate SKILL.md
Create the skill file using this template:
---
name: skill-name
description: Brief description in preferred language
---
# Skill Name
Brief overview of what this skill does.
## Reference Files (Read on Activation)
| File | Purpose |
|------|---------|
| `path/to/core-document.md` | **REQUIRED** - Core context, always read first |
| `path/to/reference.md` | Framework/theory reference |
| `path/to/folder/` | Recent logs for pattern awareness |
## Working Folder
New content goes in: `path/to/working-folder/`
## Modes
### Mode 1: [Name]
**Trigger**: "phrase that activates this mode"
**Workflow**:
1. Step one
2. Step two
3. ...
**Output**: What gets created/updated
### Mode 2: [Name]
...
## Templates
### [Template Name]
\```markdown
Template content here
\```
## Style Guide
- Principle 1
- Principle 2
- ...
## Triggers (When to Suggest This Skill)
| Signal | Action |
|--------|--------|
| User says X | Suggest this skill |
Step 7: Test
Run through a sample interaction:
- Invoke the skill
- Try each mode
- Verify it reads context correctly
- Verify it creates output in the right location
Examples
Example 1: Healing Companion
Core Document: 20_Areas/Health/焦虑型依恋-我的疗愈地图.md
Reference Files: CBT prompts, attachment theory, healing songs
Working Folder: 20_Areas/Health/疗愈日志/
Modes: Trigger processing, check-in, review, crisis, songs
Style: Therapeutic, validating, one question at a time
Example 2: Career Coach (hypothetical)
Core Document: 20_Areas/Career/Career Development Map.md
Reference Files: Resume, skills inventory, industry research
Working Folder: 20_Areas/Career/reflections/
Modes: Goal review, interview prep, skill assessment, frustration processing
Style: Encouraging but realistic, action-oriented
Example 3: Language Learning (hypothetical)
Core Document: 10_Projects/Japanese-Learning/Learning Plan.md
Reference Files: Grammar guides, vocabulary lists, resources
Working Folder: 50_Journal/language-practice/
Modes: Lesson, practice, review, immersion log
Style: Patient, encouraging, progressively challenging
Key Principles
- Read before responding: Always read the Core Document on activation
- Write to remember: Log insights and sessions to the Working Folder
- Reference frameworks: Apply theories and methods from Reference Files
- Respect user autonomy: Ask before creating files or making updates
- Bilingual awareness: Match user's language, search in both languages
Integration with Other Skills
- vault-index: Find related notes and backlinks
- moc-manager: Create index notes for project content
- knowledge-synthesis: Connect insights across vault
#meta-skill #methodology #project-builder