Quickstart¶
New to Ralph Workflow? Start with Getting Started if you want the same flow with more explanation.
Ralph Workflow is the operating system for autonomous coding — a free and open-source CLI that runs the coding agents you already use on your own machine.
It is for developers and technical teams with work that is too big to babysit and too risky to trust blindly.
What makes it different is the loop: Ralph Workflow plans the task, builds it, verifies the result, and leaves you with finished code in your repo instead of just a transcript and a claim that the task is done.
Why use it now? Because you can run one real backlog task tonight, come back to finished code tomorrow, and ask one honest question: would I merge this?
Important first-run expectation: Ralph Workflow does not replace the coding agent itself. Before you install, have at least one supported agent CLI already installed and already authenticated on your own machine.
The fastest honest first run¶
Use this flow in a real repo you already care about:
Checklist before you start:
Python 3.12+
a git repo you can safely test in
at least one supported agent CLI already working on your machine
If you are unsure which agent to start with, use the one already installed and read Which Agent Should I Start With?. If Claude Code is already your default and the real thing you want is a more trustworthy automation / unattended handoff, read Claude Code Automation for Real Repo Work. If Claude Code approval mode or plan mode still leaves you babysitting the run, read Claude Code Approval Mode Is Not an Unattended Workflow. If Claude Code is already your default and you want the clearest reason to add Ralph Workflow, read Ralph Workflow vs Claude Code. If Codex CLI is already your default and you want to know when a reviewable unattended handoff is the better fit, read Ralph Workflow vs Codex CLI. If you already split work between Claude Code and Codex, read Claude Code + Codex Workflow. If you already run multiple agents and the trust gap is in the morning-after handoff, read What Breaks First When You Run Multiple Coding Agents?. If the merge decision itself still feels fuzzy, read How to Review AI Coding Output Before You Merge. If the missing piece is a short trustworthy re-entry summary, read What a Good AI Coding Finish Receipt Looks Like. If the missing piece is keeping unattended runs bounded instead of letting them drift, read Bounded Autonomy for Unattended Coding. If the thing you keep calling remote supervision is really a finish-state trust problem, read Remote Supervision of Coding Agents. If you want the Codeberg-first category explanation before you install, read Open-Source AI Coding Orchestrator: What Ralph Workflow Is Actually For. If you are evaluating orchestration tools directly, read AI Agent Orchestration CLI: A Practical Comparison for Developers. If you want the spec-first framing before the first run, read Spec-Driven AI Agent: Why the Spec Matters More Than the Prompt.
Install¶
pipx install ralph-workflow
ralph --version
Initialize Ralph Workflow in a repository¶
Go to your project directory, then run:
cd <your-project>
ralph --init
This creates:
PROMPT.md— the task file in the project root.agent/— project-local support files (mcp.toml,pipeline.toml,artifacts.toml)~/.config/ralph-workflow.tomland~/.config/ralph-workflow-mcp.toml— user-global defaults created once and reused across projects
If this repository also needs a project-local copy of the main Ralph Workflow config, run the explicit opt-in local-override flow:
ralph --init-local-config
That command creates .agent/ralph-workflow.toml as the project-local main-config override.
Edit PROMPT.md¶
Open PROMPT.md and replace the example with one real, bounded backlog task. If you are unsure what a good first task looks like, read Choose Your First Ralph Workflow Task first. If you want copy-paste prompt shapes instead of starting from a blank page, read First-Task Prompt Templates.
A strong first prompt looks like this:
# Goal
Add validation so the CLI rejects empty project names before creating files.
Keep the rest of the flow unchanged.
## Acceptance criteria
- Empty or whitespace-only project names fail with a clear error
- No project files are created for invalid names
- Existing valid-name behavior stays unchanged
- Tests cover the new validation
Important: remove the <!-- ralph:starter-prompt ... --> comment at the top after replacing the example content. Ralph Workflow refuses to run while that sentinel is still present.
Verify the environment¶
ralph --diagnose
The diagnostic checks the repo, config, agent binaries, MCP definitions, and prompt pre-flight state. Fix any ❌ rows before running.
Run Ralph Workflow¶
ralph
Ralph Workflow runs unattended and shows progress inline. In plain terms, it plans the task, builds it, verifies the result, and leaves you with finished code to judge.
If the run earns your trust, put the public signal on the primary Codeberg repo first: https://codeberg.org/RalphWorkflow/Ralph-Workflow. The GitHub mirror stays available here: https://github.com/Ralph-Workflow/Ralph-Workflow.
If interrupted, Ralph Workflow saves a checkpoint automatically. Continue from that saved state with:
ralph --resume
How to judge the result honestly¶
Do not ask whether the agent sounded convincing.
Ask:
does the diff match the task?
did the checks really run?
are the changes reviewable in one sitting?
would I merge this?
That is the real first-run test.
If you want the shortest post-run scorecard plus the right public next step, use After Your First Ralph Workflow Run.
Where to go next¶
Choose Your First Ralph Workflow Task — pick a real first task with a clean pass/fail evaluation
Which Agent Should I Start With? — pick the first agent path with the least setup friction
Claude Code Automation for Real Repo Work — the clearest Codeberg-first path if your current problem is specifically Claude Code automation
Claude Code Approval Mode Is Not an Unattended Workflow — the clearest path if approval or plan mode still leaves you babysitting instead of reviewing a clean morning-after handoff
First-Task Prompt Templates — copy-paste starter specs for common good-fit tasks
Ralph Workflow vs Claude Code — the clearest comparison if your baseline is an interactive Claude Code session
Ralph Workflow vs Codex CLI — the clearest comparison if your baseline is an interactive Codex CLI session
Claude Code + Codex Workflow — practical guide for reducing manual handoff glue between implementation and review
What Breaks First When You Run Multiple Coding Agents? — practical guide for the trust/reconstruction failures that show up before raw Git conflicts
How to Review AI Coding Output Before You Merge — practical five-minute merge checklist for the morning-after handoff
What a Good AI Coding Finish Receipt Looks Like — the exact short handoff that should tell you what changed, what passed, and what still needs judgment
After Your First Ralph Workflow Run — quick post-run scorecard plus the right Codeberg-first next step
Bounded Autonomy for Unattended Coding — keep unattended runs fail-closed, cheap to miss, and easy to inspect the next morning
Ralph Workflow vs Aider — the clearest comparison if your current baseline is interactive AI pair programming
What Good Ralph Workflow Output Looks Like — see the shorter proof overview and the merge test
Example Review Bundle — inspect a public sample prompt, result notes, review feedback, and artifacts before your own first run
Getting Started — fuller first-run walkthrough
Concepts — terminology and mental model
CLI Reference — all flags and sub-commands
Configuration Reference — config files and precedence
Python API Reference — package documentation