@klein-kit/orchestrate

Cross-agent workflows for local coding agents. Chain the agents on your machine — Claude plans, Codex implements, an independent agent reviews — with dependencies, retries, and automatic state handoff. Built on @klein-kit/core.

import { workflow, runWorkflow } from "@klein-kit/orchestrate";

const def = workflow("dark-mode", "Add a dark-mode toggle to settings")
  .plan("plan", { agent: { provider: "claude" } })
  .implement("impl", { agent: { provider: "codex" }, dependsOn: ["plan"] })
  .verify("tests", "pnpm test", { retries: 2 })      // failing tests → repair agent → re-run
  .review("review", "impl", { agent: "auto", retries: 1 })
  .build();                                          // plain JSON — serialize, persist, ship

const result = await runWorkflow(def, { onEvent: (e) => render(e) });

• Task-state handoffs — a structured document (goal, plan, todos, decisions, files touched, per-agent history with resumable session ids) kept current from native events and injected at each handoff.
• Reviewer loops — VERDICT: REVISE re-runs the target with the reviewer's notes; exhausted reviews fail the workflow.
• Verify/repair — a failing command sends its output to a repair agent, then re-runs (CI repair).
• Auto agent selection — picks from installed providers per role; reviewers are always a different provider than the implementer.
• Conflict reconciliation — concurrent steps touching the same files are detected and reconciled by an agent.
• Subagents on every provider — runSubagent() gives scoped child sessions even where the backend has none.

Full docs: github.com/yvescomputer/klein-kit · MIT © Yves Computer Company