Why workflows matter
In a typical AI coding tool, you are the orchestrator. You prompt, read the output, decide the next step, prompt again. For simple tasks that’s fine. For anything involving multiple steps — refactoring, code review, test-driven development, release prep — you’re doing the same manual choreography every time. Workflows automate that choreography. You define the steps, the agents, and the logic once. Then you run it.- Without workflows
- With a Reliant workflow
What ships with Reliant
Reliant includes built-in workflows you can use immediately:| Workflow | What it does |
|---|---|
| agent | The default. Chat with an AI agent that can read, edit, search, and run commands in your project. |
| gsd | ”Get Shit Done.” Discuss → research → plan → implement in parallel waves → verify. Fresh context per phase. |
| spec-driven | Spec-first development. Write requirements → plan technical approach → implement. |
| simplify-first | Refactor before building. Research simplifications → refactor → verify → plan → implement. For messy or legacy code. |
| superpowers | Mandatory TDD. Brainstorm → plan → write failing tests → implement until green → structured review. |
| get-it-right | For complex brownfield codebases. Expects first attempts to fail — attempt → evaluate → improve or restart → diagnose → final implementation. |
| one-ring | Full pipeline: planning → testing → implementation → checks → code review → evaluate → refactor loop. |
| parallel-compete | Three agents implement the same thing in isolated worktrees. A reviewer picks the winner. |
| router | Classifies your request and routes it to the best workflow automatically. |
What you can build
Beyond the built-ins, workflows let you encode any repeatable multi-step process: Automated code review — Parallel agents check security, style, and correctness. A third agent synthesizes findings into one review. Test-driven bug fixes — Write a failing test, loop: attempt fix → run tests → retry if failing. Stops when tests pass or hits an attempt limit. Refactor with guardrails — Analyze code for simplification opportunities, refactor, then verify that tests still pass before moving on. Competitive implementations — Three agents solve the same problem with different approaches in isolated worktrees. A reviewer picks the best one. Plan-debate — A proposer and critic alternate rounds, refining a plan until they converge. Better plans through adversarial thinking. Continuous autonomous development — Pick a task from a spec, implement it, verify it, commit it, pick the next task. Hands-off until done.Config as code
Workflows are YAML files that live in.reliant/workflows/ inside your repository. This gives you things no other ADE offers:
- History — Track who changed what workflow and when
- Review — Workflow changes go through PRs like any other code
- Consistency — Everyone on the team runs the same processes
- Rollback — Revert a broken workflow like any other commit