For AI agents · 48 patterns

Use the library from your coding agent

Every card is also published in formats coding agents can read. Point your agent at the library and it can pick the right patterns, read their failure modes and design around them. All files are generated from the same source as these pages, so they're always current.

Files

What's available

llms.txt · ~9 KBIndexEvery pattern with its one-line intent and link. Start here. llms-full.txt · ~200 KBFull textEvery card as Markdown in one file: problem, solution, trade-offs, failure modes, example. patterns.json · ~220 KBStructured dataAll fields per pattern, for tools and scripts. IDs are stable. SKILL.md · ~22 KBAgent skillUsage rules plus an index with "use when" for every pattern.

Option 1

Install the skill

For coding agents that support skills, such as Claude Code, save the skill into your repo. The agent loads it when a task involves designing an agent or setting up how coding agents work, and fetches full cards when it needs them.

PowerShell

New-Item -ItemType Directory -Force .claude/skills/agentic-patterns | Out-Null
Invoke-WebRequest https://vikasgoyal.github.io/agentic/patterns/skill/SKILL.md -OutFile .claude/skills/agentic-patterns/SKILL.md

bash / zsh

mkdir -p .claude/skills/agentic-patterns
curl -fsSL https://vikasgoyal.github.io/agentic/patterns/skill/SKILL.md -o .claude/skills/agentic-patterns/SKILL.md

For other tools, use the folder they read skills from. Re-run the command to update.

Option 2

Add a pointer to AGENTS.md or CLAUDE.md

Works with any coding agent that reads a project instructions file and can fetch web pages.

## Agentic design patterns
When designing or reviewing an agent, a multi-agent system or an agentic feature,
or when changing how coding agents work in this repo, consult the Agentic Pattern Library:
- Index: https://vikasgoyal.github.io/agentic/patterns/llms.txt
- Full cards: https://vikasgoyal.github.io/agentic/patterns/llms-full.txt
Name the patterns you apply and why, and handle each pattern's failure modes.
Values in the worked examples are illustrative; derive real ones from our requirements.

Option 3

Use it in a single prompt

Read https://vikasgoyal.github.io/agentic/patterns/llms.txt.
For the design below, pick the 2–4 patterns that fit best, explain why,
and list how the design handles each pattern's failure modes
(full cards: https://vikasgoyal.github.io/agentic/patterns/llms-full.txt).

Design: <describe the agent or feature>

Notes

Good to know

Numbers in worked examples, like thresholds, caps and timings, are illustrative. The skill tells agents not to copy them.
Pattern IDs such as fast-slow-escalation are stable. New patterns are added; existing IDs won't be renamed.
patterns.json includes a generated timestamp, so tools can tell when the library last changed.