Getting Started

Ten minutes from install to running, testing, and gating intent. Everything here is deterministic and runs with no AI.

1. Install

npm install -g @skillstech/intentlang
intent help

2. Scaffold a mission

intent init Eligibility

That writes Eligibility.intent, a real starter with a goal, a guarantee, a never rule, an executable decision, and in-file test cases. It is valid and runnable out of the box.

3. Run it (no code, no AI)

A decision is a program. Give it inputs and it decides:

intent run Eligibility.intent --inputs '{"age":20}'
#   decision Example: Allowed  [rule: adult]

The trace shows which rule fired. Change the input and the result changes, deterministically, before any implementation exists.

4. Test it, in the same file

The test blocks assert behavior through the same runtime:

intent test Eligibility.intent
#   intent test Eligibility.intent: 2/2 passed

The .intent file is now self-verifying. No test framework, no code.

5. Format it

intent fmt Eligibility.intent --write

Canonical whitespace, comments preserved. Use intent fmt . --check in CI to keep a whole tree consistent.

6. Check it

intent check Eligibility.intent

Diagnostics for one file, or gate a whole repo at once:

intent check .        # recurses every .intent, exits non-zero on any error

Add it to CI with the GitHub Action:

- uses: SkillsTechTalk/intent-language@main
  with:
    paths: ./intent

7. Edit with intelligence

Install the editor support: the Language Server (intent lsp) gives live diagnostics, completion, and hover in VS Code, Neovim, and any LSP editor, plus syntax highlighting via the shipped grammar.

8. Go further

• Executable intent in depth: the Intent Runtime and first-class tests.
• Interop: export to DMN/BPMN/JSON-Schema/OpenAPI and import back.
• The whole language: the syntax overview and the specification.
• The model everyone builds on: Intent for every role.
• Already have a codebase? Adopt IntentLang on it: lift, review, check, gate, and keep in sync , one mission at a time.

The whole loop, author, run, test, format, gate, in one deterministic toolchain. That is what beyond prompt engineering looks like in practice.