agentic-patterns-source
npx skills add https://github.com/laurigates/claude-plugins --skill agentic-patterns-source
Agent 安装分布
Skill 文档
Agentic Patterns Source
Overview
Awesome Agentic Patterns is a curated catalog of production-ready AI agent patterns. It bridges the gap between toy demos and production systems by documenting repeatable solutions that teams are actually using.
Primary URL: https://agentic-patterns.com/
When to Use This Skill
| Use this skill when… | Use other skills when… |
|---|---|
| Designing multi-agent orchestration | Implementing Claude Code-specific hooks (claude-hooks-configuration) |
| Solving context window limitations | Writing custom agent definitions (custom-agent-definitions) |
| Implementing feedback/self-correction loops | Creating file-based coordination (agent-file-coordination) |
| Researching security/sandboxing patterns | Setting up MCP servers (mcp-management) |
| Need production-tested patterns with references | Need Claude Code workflow patterns (multi-agent-workflows) |
Pattern Categories
1. Context & Memory
Managing limited context windows through curation, caching, and episodic memory.
Key Patterns:
- Context Window Auto-Compaction
- Progressive Disclosure for Large Files
- Semantic Context Filtering
- Self-Identity Accumulation
When to research: Context limits affecting performance, memory persistence needs, large codebase navigation.
2. Feedback Loops
Self-healing mechanisms, CI integration, and iterative refinement.
Key Patterns:
- Self-correction loops
- CI integration patterns
- Iterative refinement workflows
When to research: Implementing retry logic, self-healing agents, continuous improvement cycles.
3. Learning & Adaptation
Reinforcement fine-tuning and skill library evolution.
Key Patterns:
- Memory Reinforcement Learning (MemRL)
- Skill Library Evolution
When to research: Agents that improve over time, skill accumulation, adaptive behavior.
4. Orchestration & Control
Task decomposition, multi-agent coordination, and tool routing.
Key Patterns:
- Planner-Worker Separation for Long-Running Agents
- Lane-Based Execution Queueing
- Custom Sandboxed Background Agent
- Tool routing patterns
When to research: Complex multi-step tasks, parallel agent execution, long-running operations.
5. Reliability & Eval
Testing harnesses, logging, and reproducibility safeguards.
Key Patterns:
- Failover-Aware Model Fallback
- Testing harnesses
- Observability patterns
When to research: Production reliability, testing agent behavior, debugging failures.
6. Security & Safety
Sandboxing, PII protection, and deterministic scanning.
Key Patterns:
- External Credential Sync
- Sandboxed Tool Authorization
- PII protection patterns
When to research: Secure agent execution, credential management, data protection.
7. Tool Use & Environment
Shell integration, browser automation, and API design.
Key Patterns:
- Intelligent Bash Tool Execution
- Browser automation patterns
- API design for agents
When to research: Tool integration, shell command patterns, API consumption.
8. UX & Collaboration
Human handoffs, async workflows, and transparency mechanisms.
Key Patterns:
- Human-in-the-loop patterns
- Async workflow management
- Transparent reasoning displays
When to research: User interaction design, approval workflows, explainability.
Research Workflow
Finding Patterns for a Specific Challenge
-
Search the catalog
WebSearch: site:agentic-patterns.com {challenge_keywords} -
Fetch pattern details
WebFetch: https://agentic-patterns.com/patterns/{category}/{pattern-name} Prompt: Extract the pattern description, implementation guidance, and referenced examples -
Cross-reference with local skills
- Compare with existing agent-patterns-plugin skills
- Identify gaps or enhancement opportunities
Exploring a Category
WebFetch: https://agentic-patterns.com/
Prompt: List all patterns in the {category} category with brief descriptions
Staying Current
The site actively maintains patterns with NEW and UPDATED badges:
-
Check for new patterns
WebFetch: https://agentic-patterns.com/ Prompt: List patterns marked as NEW or UPDATED in the past month -
Review for applicability
- Does this pattern address a gap in our plugin?
- Can we adapt it to Claude Code workflows?
- Should we create a new skill based on this pattern?
Pattern Applicability to Claude Code
| Site Category | Claude Code Plugin/Skill |
|---|---|
| Context & Memory | Context management in CLAUDE.md, skill loading |
| Feedback Loops | Pre/post hooks, test-driven workflows |
| Orchestration & Control | multi-agent-workflows, delegation-first |
| Tool Use & Environment | claude-hooks-configuration, mcp-management |
| Security & Safety | Permission systems, tool restrictions |
| UX & Collaboration | agent-handoff-markers, approval workflows |
Integration with Other Skills
This skill complements:
- multi-agent-workflows: Apply catalog patterns to Claude Code agent design
- agent-coordination-patterns: Enhance coordination with production patterns
- delegation-first: Research delegation patterns from real implementations
- custom-agent-definitions: Use patterns to inform agent specialization
Delegation Pattern
For deep pattern research:
Use research-documentation agent for:
- Comprehensive pattern analysis
- Cross-referencing multiple patterns
- Building implementation recommendations
Example delegation prompt:
"Research agentic-patterns.com for patterns related to {challenge}.
Extract implementation guidance and compare with our existing
agent-patterns-plugin skills. Identify patterns we should adopt."
Agentic Optimizations
| Context | Approach |
|---|---|
| Quick pattern lookup | WebSearch with site filter |
| Category exploration | WebFetch main page, filter by category |
| Deep pattern analysis | Delegate to research agent |
| Staying current | Monthly review of NEW/UPDATED patterns |
Quick Reference
| Need | Search Query |
|---|---|
| Context limits | site:agentic-patterns.com context window compaction |
| Multi-agent | site:agentic-patterns.com orchestration multi-agent |
| Self-healing | site:agentic-patterns.com feedback loop self-correction |
| Security | site:agentic-patterns.com sandbox authorization |
| Tool use | site:agentic-patterns.com tool execution shell |