Cookbook

Each recipe is self-contained — a complete experience.yaml snippet you can drop in and adapt. For the full feature set, see Concepts and the YAML schema reference.

Recipes

Control flow

Recipe	What it covers
Fan-out with concurrency	for_each over an array with concurrency: N; merge: array_append
Branch by feature flag	when: edges to route at runtime based on a state value
Retry with backoff	on_error: { policy: retry } with linear and exponential backoff
Nested experiences	kind: experience node; state_scope: isolated; passing args

State & output

Recipe	What it covers
Merge strategies	array_append / set_once / last_wins with worked examples
Structured output schemas	schema: patterns — nested objects, enums, arrays, nullable fields

Running & recovery

Recipe	What it covers
Resume from cache	oe resume <run-id>; cache invalidation; --from flag

Visualize & share

Recipe	What it covers
Visualize a graph & share a run report	oe graph Mermaid DAG; oe inspect --html self-contained run report

CI / automation

Recipe	What it covers
Run on every PR & pin the report	pull_request workflow + reusable oe-report composite action; sticky PR comment + uploaded oe inspect --html artifact; pull_request vs pull_request_target security

Multi-provider & integration

Recipe	What it covers
CLI agent with file edits	cli-agent with allow_file_edits: true; structured output
Hybrid LLM routing	Anthropic + OpenAI in one flow via meta.runtime.providers
oe-mcp as server	Expose OE flows as MCP tools for Claude Code / Codex / Gemini
MCP resource as a dataset source	source.type: mcp-resource; mcp.json config; row normalization
Cross-vendor CLI agent chain	Claude Code → Codex → Gemini in series; each reads prior output

Registry & publishing

Recipe	What it covers
Submit to the registry	oe submit; auto-detected GitHub metadata; pre-submit checklist

---

More patterns live in the examples gallery — every bundled example is a runnable recipe.

Stuck? Check FAQ + troubleshooting.