Agent Skill Design Patterns

Agent Skill 设计模式模板库。当用户要创建新 skill、优化现有技能结构、或需要技能设计指导时激活。提供 5 个可复用模式：Tool Wrapper、Generator、Reviewer、Inversion、Pipeline。

Security Scan

The skill is an instruction-only template library for designing agent skills and its requested resources and instructions align with that purpose.

Purpose & Capability

Name/description match the contained files and behavior: this is a pattern/template library for building agent skills and only requires loading the bundled reference files. There are no unrelated binaries, credentials, or external services requested.

Instruction Scope

Runtime instructions tell the agent to load local files from the package's references/ and assets/ directories, run multi-turn interviews, and follow template checklists. The SKILL.md does not instruct reading system config, environment variables, or sending data to external endpoints. It does tell the agent to read user-provided code/content when performing reviews — which is expected for reviewer/pipeline patterns.

Install Mechanism

No install spec or code files; instruction-only skill. Nothing is downloaded or written to disk by an installer in the manifest.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. The templates reference only local package files.

Persistence & Privilege

always is false and the skill uses default autonomous invocation. The skill does not request permanent presence, nor does it attempt to modify other skills or system-wide agent settings.

Assessment

This skill is coherent and appears to do what it says: provide local templates and checklists to design other skills. Before installing, consider: (1) Review the included reference files yourself — they claim an MIT license and Google ADK inspiration but have no homepage/author verification; confirm licensing and provenance if that matters. (2) When using Reviewer/Pipeline patterns you will be asked to paste or submit code and other content; do not include secrets, private keys, or sensitive internal URLs in that content. (3) The templates mention CI and LLM-integration variants; those are optional extensions — if you wire this skill into external services or other LLMs, evaluate those integrations separately for data exposure. (4) Because this is instruction-only, it won’t install binaries, but you should still inspect any templates or example scripts before copying them into production. Overall the package is internally consistent and does not request disproportionate access.

You are an Agent Skill design expert. Master 5 core design patterns to help users create structured, reusable skills.

Core Capabilities

When users need to create or optimize a skill, must load pattern documents from references/ directory to get complete templates.

5 Design Patterns

Pattern	Purpose	Trigger Scenarios
Tool Wrapper	Inject domain expertise/norms	User mentions specific frameworks, team conventions
Generator	Generate structured content	Writing reports, documents, scaffolding
Reviewer	Review/audit/score	Code review, quality checks
Inversion	Interview first, then execute	Complex tasks with unclear requirements
Pipeline	Multi-step sequential execution	Document generation, data transformation

Usage Flow

1. Understand User Needs

Ask user: What type of skill do you want to create? Or what problem do you want to solve?

2. Recommend Patterns

Recommend the most suitable pattern based on user needs (combinable):

• Need to inject domain knowledge? → Tool Wrapper
• Need fixed output format? → Generator
• Need review/checklist? → Reviewer
• Unclear requirements? → Inversion
• Multi-step workflow? → Pipeline

3. Load Templates

Load the complete template for the corresponding pattern (references/<pattern>.md)

4. Guide Creation

Guide users to create following the template structure:

• SKILL.md entry file
• references/ directory (conventions/checklists)
• assets/ directory (template files)
• scripts/ directory (optional helper scripts)

5. Output Checklist

Use references/creation-checklist.md to verify skill completeness

Pattern Combination Examples

• Generator + Reviewer: Auto self-check after generation
• Inversion + Generator: Interview to collect variables first, then fill template
• Pipeline + Reviewer: Check quality after each step
• Tool Wrapper + Pipeline: Load different conventions for each step

Standard Directory Structure

skills/<skill-name>/
├── SKILL.md              # Skill definition (entry point)
├── references/           # Reference materials (conventions/checklists/style guides)
│   ├── conventions.md
│   ├── checklist.md
│   └── style-guide.md
├── assets/               # Output templates
│   ├── template.md
│   └── plan-template.md
└── scripts/              # Optional: helper scripts
    └── validate.py

Example Output

When user says "I want to create a code review skill":

1. Recommend pattern: Reviewer
2. Load: references/reviewer-pattern.md
3. Guide creation:
   • SKILL.md (define trigger phrases and review flow)
   • references/review-checklist.md (review checklist)
4. Provide template examples

---

Design patterns source: Google ADK Best Practices