# Master Claude Skills (Module 2: Architecture)
**作者**: HereIsYourAi
**日期**: 2026-04-17T10:13:11.000Z
**来源**: [https://x.com/Ai_here202/status/2045083140008579535](https://x.com/Ai_here202/status/2045083140008579535)
---
In Module 1, you learned how to build and deploy your first Skill.
Now comes the real shift:
> Moving from single-use Skills → to a scalable system of intelligent automation
Because once you’ve built a few Skills, new challenges appear:
- Conflicts between Skills
- Tasks that require computation, not just instructions
- Managing large references efficiently
This module solves all of that.
## ⚙️ When Instructions Are Not Enough
So far, everything you’ve built relies on plain English instructions.
That works well for:
- Writing
- Formatting
- Categorization
- Decision-making
But some tasks go beyond language.
They require:
- Calculations
- File manipulation
- Data processing
- External integrations
That’s where scripts come in.
🧩 Introducing the scripts/ Layer
Think of your Skill as having two components:
- Instructions → Handle reasoning and decisions
- Scripts → Handle execution and computation
## 📌 When Should You Use Scripts?
Use Instructions When:
- The task involves judgment
- Output depends on interpretation
- Language quality matters
Examples:
- “Rewrite this in a professional tone”
- “Summarize meeting notes”
- “Categorize feedback”
Use Scripts When:
- Precision is critical
- Computation is required
- Files need to be processed
Examples:
- Parsing CSV or XML files
- Performing calculations
- Resizing images
- Extracting structured data
Use Both When:
Some workflows need both logic and execution.
Example:
> Process a dataset (script) → then generate a human-readable summary (instructions)
## 🔗 How Scripts Work Inside a Skill
Your Skill structure now evolves:
your-skill/
├── SKILL.md
├── scripts/
│ └── process-data.py
└── references/
Key Insight:
> Scripts do the work.
> Instructions decide when and how to use them.
## 🧠 Script Best Practices
To keep your system reliable and scalable:
1. One Script = One Responsibility
Avoid combining multiple tasks in one script.
✔️ Good:
- parse-csv.py
- calculate-metrics.py
❌ Bad:
- A single script doing everything
2. Use Arguments (Not Hardcoding)
Your scripts should accept inputs dynamically:
- Input file path
- Output file path
This keeps your Skill flexible.
3. Add Error Handling
Your script should clearly communicate failures:
- Missing file
- Incorrect format
- Invalid data
Claude can read these errors and respond intelligently.
4. Document the Script
At the top of every script, include:
- What it does
- Required inputs
- Expected outputs
- Possible errors
This ensures clarity—for both you and the system.
🔄 Multi-Skill Orchestration
Once you create multiple Skills, a new problem emerges:
> Which Skill should run?
## ⚠️ The Conflict Problem
Example:
- You ask: “Draft an email”
- But your “Content Repurposer” Skill activates
Or:
- Two Skills respond to the same trigger
This creates inconsistency and friction.
## 🧠 How Claude Chooses a Skill
Behind the scenes:
Claude reads all Skill descriptions
It compares them to your request
It assigns relevance scores
The highest-scoring Skill activates
If none meet the threshold → nothing runs
## 🚨 Why Conflicts Happen
- Overlapping trigger phrases
- Vague descriptions
- Missing boundaries
✅ Rules for Multi-Skill Harmony
To build a clean, scalable system:
## 1. Define Clear Territories
Each Skill must have a specific domain
✔️ Example:
- Email Drafter → writes emails
- Brand Voice Enforcer → checks tone
- Content Repurposer → transforms format
No overlap.
## 2. Use Strong Negative Boundaries
Explicitly define exclusions:
- “Do NOT use for email drafting”
- “Do NOT use for tone checking”
This prevents misfires.
## 3. Create Distinct Trigger Language
Each Skill should have unique activation phrases
✔️ Example:
- “Draft an email” → Email Skill
- “Check tone” → Brand Skill
If two Skills share triggers → you have a scope issue
## 🧪 Diagnosing Conflicts
If the wrong Skill activates:
- Review the YAML description
- Identify overlapping phrases
- Tighten scope and exclusions
In most cases, the issue is not the logic—it’s the description.
📚 Advanced Reference Strategies
In Module 1, you learned basic reference handling.
But real-world use cases are more complex.
## ⚠️ The Problem with Large References
What if your Skill needs:
- A 50-page brand guide
- A 30-page style manual
- Multiple templates
If you load everything at once:
> You waste context and reduce performance
## 🧠 Smart Reference Design
Instead of loading everything blindly:
- Structure references logically
- Only access what’s needed
- Keep files concise and focused
## 📌 Key Principle:
> The more selective your references, the better your Skill performs.
🏗️ Building Scalable Architecture
At this stage, your system should look like:
- Multiple specialized Skills
- Clear boundaries between them
- Scripts handling computation
- References optimized for efficiency
## 🔑 The Big Shift
You’re no longer just “using Claude”
You’re:
> Designing a modular system of intelligent workflows
## 相关链接
- [HereIsYourAi](https://x.com/Ai_here202)
- [@Ai_here202](https://x.com/Ai_here202)
- [process-data.py](https://process-data.py/)
- [parse-csv.py](https://parse-csv.py/)
- [calculate-metrics.py](https://calculate-metrics.py/)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [6:13 PM · Apr 17, 2026](https://x.com/Ai_here202/status/2045083140008579535)
- [8,090 Views](https://x.com/Ai_here202/status/2045083140008579535/analytics)
---
*导出时间: 2026/4/18 17:39:04*