Use the catalog from an agent
Every pattern has a stable slug and a Markdown twin. All endpoints below are static files. You do not need an API key.
Endpoints
| URL | Contents |
|---|---|
/llms.txt | Index of every pattern: slug, title, one-line summary, page URL, Markdown URL. |
/llms-full.txt | Full text of every pattern in one file. |
/patterns/action-selector-pattern.md | One pattern as Markdown with YAML front-matter. Replace the slug for any pattern. |
/patterns.json | Metadata for all patterns: title, status, category, tags, summary, signals. |
/data/patterns/action-selector-pattern.json | Metadata for one pattern as JSON. |
/graph.json | Nodes and "related" edges between patterns. |
/rss.xml | Feed of added and updated patterns. |
From the command line
# Find patterns about prompt injection
curl -s https://agentic-patterns.com/llms.txt | grep -i -B1 -A2 injection
# Read one pattern
curl -s https://agentic-patterns.com/patterns/action-selector-pattern.mdFrom a coding agent
Give your agent a prompt like this one:
Read https://agentic-patterns.com/llms.txt and pick the patterns that fit this project.
For each one, fetch its .md file, compare its "Use when" and "Avoid when"
lists with the codebase, and tell me which pattern to apply first and why.On any pattern page, Copy for agent copies the pattern as Markdown with a short instruction to check fit and list the files to change first.
Discovery
Every page links /llms.txt with <link rel="alternate">.
Pattern pages also link their Markdown twin with type="text/markdown", so a
browsing agent can find the Markdown from the HTML page.
Something missing?
Tell us what your agent needs in a GitHub issue.