RabbinatixDocs
Sign inGet started

Work as a team

AI assistant or CLI: which does what

Where your AI assistant works and where the rabbinatix CLI works, the rules between them, and every feature that touches code.

Open AI assistants in the app

The rule

Rabbinatix works with your code in two ways: through your AI assistant, in your editor, and through the rabbinatix CLI, in your terminal and your CI. Each has its own job.

QuestionYes: use
Does it need understanding or judgement: reading code for its meaning, writing code, changing the map, explaining, choosing a fix?The AI assistant
Must it run without a person, give the same answer every time, block something (a merge, a deployment), or run without sending code anywhere?The CLI
Both?The assistant creates it (once, with you); the CLI checks it (on every commit, without you).

Side by side

AI assistantCLI (rabbinatix)
What it isYour own assistant (Claude Code, Cursor, Copilot and others) connected to RabbinatixAn npm package: npx -y rabbinatix <command>, Node 20 or later, on Linux, macOS and Windows
Who runs itYou, in your editorYour CI on every pull request and deploy, or you in a terminal
How it decidesIt reads and reasons, so two runs can differFixed rules: the same input gives the same answer
What it may change in RabbinatixDrafts only (you accept them), the code findings it reports, and which it fixedRecords only: deployments, impact reports, code findings, watcher reports. Never your design
What it may change in your repoCode, anchors, tests, as you askNothing, except the files you ask a command to write
How it connectsAn assistant key (read, draft, report)An API token (read, deploy, scan)

The rules between them

  1. The CLI never calls an AI model. Where a result needs understanding, it prints the AI prompt to give your assistant instead.
  2. The assistant is never a gate. Nothing in CI waits on an AI model's answer. The assistant can run the CLI to get a fixed one.
  3. Only you and your assistant change the design: systems, APIs, logic, flows and test cases. The CLI only records what happened.
  4. Your code doesn't leave your machine through the CLI. Each command says exactly what it sends.
  5. The anchors are the bridge: the assistant writes FDD: anchors in your code, and the CLI reads them.

Every feature

"—" means it isn't that side's job.

FeatureAI assistantCLIIn the app
Put a repo on the mapReads the code and drafts systems, APIs, calls, logic and flows; adds the anchors once you accept themrabbinatix status: route files with no anchor yet, and logic with no codeYou accept the drafts
Keep code in step with logicUpdates the code a logic change touchesrabbinatix status: anchors behind the map (r9 in the code, r11 on the map); --fail-on behindThe out-of-step note
Write code from logicWrites it in your project's own style and languagerabbinatix generate: the TypeScript, Python, C#, Java, Go, Kotlin, Swift, Dart, PHP, Ruby or Rust files the Code panel gives; --check fails when the repo's are olderThe Code panel
Verify generated code—rabbinatix verify: runs the Python, C#, Java, Go, Kotlin, Swift, Dart, PHP, Ruby or Rust build against recorded test runs on your machineTypeScript is verified in the browser
Code impactBefore editing (what will this affect?) and after (fixes what the change caused); keeps the report when you askrabbinatix impact: the pull request check and comment, and any two branches or commits (--base, --head)The Code impact page
Security: the designRuns the checks, explains each finding and drafts the fixes—Security checks
Security: rule-based code scan (secrets, risky code, vulnerable dependencies)—rabbinatix scan on every pull requestCode scan
Security: what needs understanding (missing access checks, logic flaws)Reads the code and reports what it finds—Code scan, marked as found by an assistant
Fix code findingsApplies the fixes you mark and says which commit fixed eachThe next scan confirms themApply fix, Dismiss, tickets
Write test casesDrafts them—You accept them
Run test casesWhile it worksrabbinatix test, or only those a change affects with --affectedRun
Safety checkExplains why not; never decidesrabbinatix deploy check (exit 1 when blocked)Deployments
Record a deployment—rabbinatix deploy recordDeployments
Test server—rabbinatix mockThe hosted test server
Test APIs—rabbinatix relay: sends from your computer, so CORS doesn't applyTest APIs (Send through → This computer)
OpenAPI out—rabbinatix openapi pull, for client generators in any languageExport
OpenAPI in step with the mapUpdates the map from a changed spec, as draftsrabbinatix openapi diff: what the spec would breakImport
Contract watcher—rabbinatix watch listen for phone apps and OpenTelemetry; rabbinatix watch import for a HAR recording or a trace fileThe watcher in your frontend
Turn a call into an expectationAny language—Generate from code (curl, fetch, axios, HttpClient)
Logic and flows: write, change, explainDrafts and explains—The editors
Questions about your systemsAnswers them—Explorer
TicketsWorks on them and plans them—The board
Build a screen from a mock-upDrafts the screen and the logic its buttons run, then checks its draft against your picture (design with draft) before you review it—The Screen Builder (AI prompts copies the prompt)
Build a screen pixel-perfectBuilds it in your project's components from the design (design, which sends the picture over MCP)rabbinatix design: the page, theme tokens, picture and every widget's box in a folderInspect in the Screen Builder and on whiteboards

How they work together

A typical change goes through both:

  1. You ask your assistant to build or change something. It drafts the logic, the flow or the code, and the anchors that tie the code to the map. You accept the drafts in Rabbinatix.
  2. You open a pull request. CI runs the CLI: rabbinatix scan, rabbinatix impact, rabbinatix status and rabbinatix test --affected. Each gives a fixed answer and can block the merge.
  3. Where the CLI finds something that needs understanding (a broken contract, code that isn't on the map, a failing test case), it prints the AI prompt to give your assistant. You paste it, and the assistant fixes or maps it.
  4. On deploy, rabbinatix deploy check gates the release and rabbinatix deploy record records it.

The same AI prompts are under AI prompts on each page of the app. Connect your assistant: Connect your AI assistant. Every command: The rabbinatix CLI.