Writing Agent Relay Workflows

This is the opinionated playbook for building multi-agent workflows with @relayflows/core. It enforces a strict contract: workflows should repair before they fail, not just discover errors and stop. Every workflow gets mandatory dual fresh-eyes loops (Claude review/fix, then Codex review/fix) with test hardening at each repair step, repairable validation gates that route failures to fix agents instead of crashing, and explicit blocked-state artifacts when real blockers appear. The checklists are prescriptive and the repair-first philosophy runs deep. If you're wiring up Claude, Codex, Gemini, Aider, or Goose into DAGs and want self-healing pipelines instead of fragile fail-fast chains, this gives you the pattern language and non-negotiables.

Install to Claude Code

npx -y skills add agentworkforce/relay --skill writing-agent-relay-workflows --agent claude-code