Recipes Overview

Overview

Recipes are end-to-end walkthroughs. Each one wires several rousseau primitives — transports, cron, provider, approval policies — into a shipped, tested workflow. Copy, adjust the JIDs / tokens, and run.

Recipe index

Design patterns you'll see

Cron + transport delivery

Every recipe that produces a scheduled artefact uses rousseau cron add with --deliver-to set to a JID / user id understood by the running transport. The scheduler goroutine invokes the transport's Deliver method — the same code path the agent uses for reply messages.

rousseau cron add \
  --name daily-standup \
  --schedule '0 9 * * MON-FRI' \
  --prompt 'Summarise yesterday'\''s commits' \
  --deliver-to '447900123456@s.whatsapp.net'

Pattern-mode approver

Most production recipes use agent.approver.mode: pattern with default: deny and specific allow-rules. This lets the agent touch the filesystem where useful (reading, grepping, editing project files) while blocking destructive shell commands.

Skills for role-specific behaviour

agent.skills_dir points at a directory of Markdown files. Each file contains agentskills.io-style YAML frontmatter and a body prompt. Rousseau discovers them and composes them into the system prompt — so a "code reviewer" skill and an "SRE triage" skill can coexist without config gymnastics.

Recipe template

If you're writing a new recipe (see Community: Contributing), follow this structure:

1. Overview — the operational problem in three sentences.
2. Prerequisites — transports, providers, external services.
3. Config — copy-pastable YAML.
4. Launch — the exact commands.
5. Verification — what "working" looks like.
6. Failure modes — the top three things we've seen go wrong.
7. Related pages — cross-links.

Related pages

• Reference: Commands: cron
• Reference: Config: Agent
• Best Practices: Approval rules
• Tutorials