Skip to main content

Rapid OS

Rapid OS v3 is a Contract-Driven Engineering OS for teams that build software with coding harnesses and AI assistants.

Its goal is not to replace the agent that writes code. Its goal is to turn AI-assisted work into an explicit, traceable, reproducible, and verifiable process: understand the repository, define what must be built, compile the right context, set execution rules, declare capabilities, record evidence, and evaluate the result.

What problem it solves​

Coding agents can work fast, but they tend to suffer from three recurring problems:

  • Loss of context: they don't always know the business rules, architecture, security requirements, or prior decisions.
  • Ambiguous instructions: a prompt is not the same as a versioned, verifiable specification.
  • Lack of evidence: an agent saying "done" does not prove that tests, reviews, or checks were run.

Rapid OS introduces deterministic contracts and records around that workflow.

What exists

rapid scan builds project intelligence with evidence and provenance.

What must be built

rapid spec manages specifications and immutable revisions.

What the agent needs to know

rapid context compiles relevant, prioritized context limited by a budget.

Under which rules it can work

rapid policy and rapid run set risk, workspace, gates, and tasks.

What the harness can do

rapid harness resolves declared capabilities against the run's contract.

What actually happened

rapid evidence and rapid eval record verifiable proof and produce a deterministic verdict.

Get started in 3 steps​

You don't need to learn commands: you work with shortcuts (/rapid-…) inside your coding agent, and in your terminal you only install and approve.

1. Install Rapid OS (Installation). At the end, the installer asks which agents to install the shortcuts for: type yours (for example claude). If you skipped it, run this once:

rapid shortcuts install --global --harness claude

2. Describe your project in the agent. It proposes the stack, the test command, and the documents; you decide:

/rapid-init a Python API to validate usernames

3. Ask for the work:

/rapid-feature add CSV export to the monthly report

When it finishes, the agent stops and gives you the command to approve in your terminal (for example rapid approve gate.final-verification). Then type /rapid-next to close out.

You want to…Type in your agent
Set up the project/rapid-init <description>
A new feature/rapid-feature <goal>
Fix a bug/rapid-bugfix <what fails>
Refactor without breaking tests/rapid-refactor <scope>
Harden security/rapid-hardening <area>
Research an option/rapid-research <question>
See where things stand or continue/rapid-status · /rapid-next

In Codex they are typed with $ ($rapid-init). In VS Code, and if you prefer to version the shortcuts with the repository, install them in the project with rapid init --shortcuts. Per-agent details are in the shortcuts guide.

Advanced mode: the v3 governance loop​

The v3 commands (rapid scan, rapid spec, rapid run, rapid evidence, rapid eval…) remain available for CI, debugging, and auditing. The v3 governance loop follows eight steps:

1. scan
2. spec
3. context
4. policy / run
5. harness
6. external execution + run update
7. evidence
8. eval + validate

Rapid OS does not autonomously run an LLM, does not approve human gates, only creates a worktree with rapid start --worktree, runs only the configured test_command or the argv of rapid check (the plumbing runs nothing), and does not use an LLM as a judge. The external coding harness does the work; Rapid OS governs and verifies the artifacts around that work.

Start here​