Stop babysitting agents. Script them.

export default flow<TicketInput>(
  "software-factory",
  { budget: "$8/run" },
  async (f, input) => {
    await f
      .agent("planner", { model: "claude-sonnet-5", task: ... })
      .gate((r) => r.artifacts.includes("plan/plan.md"));

    await f
      .agent("implementer", { model: "claude-opus-5", task: ... })
      .gate((r) => r.artifacts.includes("impl/branch.txt"));

    // The tests run outside the agent.
    // The agent cannot lie about the exit code.
    await f
      .run(`git checkout $(cat impl/branch.txt) && npm test`)
      .gate((out) => /failed:\s*0/.test(out));

    // Different model. Its only job is to break the PR.
    await f
      .agent("adversary", { model: "gpt-5", task: ... })
      .gate((r) => r.artifacts.includes("adversary/clean"));

    // Deterministic step, not an agent decision.
    await f.github.createPullRequest({
      repo: input.repo,
      head: `feat/${input.ticketId}`,
      base: input.baseBranch,
      bodyPath: "impl/summary.md",
    });

    f.done("success");
  },
);