Reyn cookbook¶

End-to-end workflows. Each subdirectory is one self-contained recipe: a README.md walking through the commands, any profile / topology / skill files the recipe needs, and a transcript.example.txt showing roughly what a successful run prints.

Format¶

Every example is driven by reyn run <skill> "<input>" and/or reyn chat <agent> — no new tooling. Recipes that need an agent or a topology call the existing reyn agent new / reyn topology new commands.

Status legend¶

✅ Works today — every dependency exists in this repo; runnable as-is.
ℹ️ Works today (custom skill) — runnable, but uses a skill bundled in the example dir rather than a stdlib skill. Copy it into reyn/local/ before running.
🔮 Roadmap — depends on a capability that is not yet implemented (a missing stdlib skill, MCP server config, scheduling, etc.). Documents the intended shape; not runnable as-is.

Index¶

Recipe	Status	What it shows
multi-agent-research	✅ Works today	3-agent team via `reyn agent new` + `reyn topology new --kind team`
translate-doc	ℹ️ Works today (custom workflow)	en → ja document translation; uses a bundled `translate_doc` workflow (no stdlib equivalent)
research-topic	🔮 Roadmap	Generic web research; `mcp_search` stdlib only finds MCP servers in the GitHub registry — generic web search needs an external search MCP server you've configured
write-readme	🔮 Roadmap	Generate a Reyn-style README; `skill_builder` builds new skills, not arbitrary docs — needs a dedicated doc-writer skill
weekly-summary	🔮 Roadmap	Cron-style recurring summarizer; needs scheduling / persistent state

Prerequisites (all recipes)¶

reyn CLI on PATH (pip install -e . from repo root, or your install method).
OPENAI_API_KEY exported in shell. (Local LiteLLM at localhost:4000 works too — set OPENAI_BASE_URL if you use it.)
A working directory with a reyn.yaml (or run from this repo root, which has one).

Conventions¶

Inputs shown as "..." are natural language; Reyn auto-wraps them as a user_message artifact.
Inputs shown as '{"type": ...}' are explicit JSON artifacts.
Transcripts are sketches — exact wording from the LLM will vary run to run.