Most agent workflows stop at "the code compiles." This one pushes to "the feature actually works, tested end-to-end, before commit." It's built around repair-before-failure validation gates: run tests with failOnError false, capture output, hand it to a repair agent, fix and rerun until green. Includes the full test-fix-rerun loop pattern, mandatory sequential Claude-then-Codex fresh-eyes review cycles over the final state, PGlite for in-memory Postgres testing, and deterministic reconciliation steps that keep repairable gates on the critical path even when long-running agents drop. The workflow writes BLOCKED_NO_COMMIT artifacts instead of crashing when repair budgets run out. Use this when "it compiles" is not enough and you need every merged commit to be production-ready with passing tests as proof.
npx -y skills add agentworkforce/relay --skill relay-80-100-workflow --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
wshobson/agents
github/awesome-copilot