CabraFlow · From narrated demo to automated test, in design
CabraFlow is a project in design that aims to turn a demo into an automated test: you would demonstrate your business process in the application while narrating it out loud, and the test would be written as business micro-phrases, replayable in Playwright or Robot Framework. As of 6 October 2026, six specifications exist, written on 12 August 2026; there is no code, no prototype and no demonstration yet.
The dead end of UI test maintenance
The problem CabraFlow aims to address: current approaches force a tradeoff between fragile tests and complex test authoring.
Dead Selectors & Disposable Scripts
Traditional recorders lock in CSS or XPath paths. At the first DOM change, renamed ID or layout adjustment, tests break, and time goes into repairing them instead of validating the application.
The Chasm Between Business & Code
Writing native Playwright, or any other TypeScript test framework, requires development skills. Functional experts, who know the business rules, can neither write nor review these tests.
Closed Formats & Per-Seat Licenses
Proprietary testing suites often keep tests in closed formats that are hard to version in Git or reuse in another tool, with licenses charged per seat.
Six design choices, recorded in the specifications
What CabraFlow is designed to do. These are written intentions, not available features: none is implemented yet.
Voice & Gesture Capture
The target mode is voice-driven test generation: you would browse while narrating your actions, and the flow would be written as the demonstration unfolds. Transcription would run locally (the Whisper engine is being considered). This mode will only be built after a scored trial.
Accessibility Tree
Targets would be resolved by their accessible roles and names, the way a screen reader perceives them, using Playwright's user-facing locators. The test language is designed without CSS or XPath selectors.
Business Micro-Phrases
The visible language would use your organization's own words ("Validate urgent purchase order"), created during capture, never a tool vocabulary or code.
Replay on Three Targets
A single YAML flow is designed to run on a native runner and export to a Playwright script (.spec.ts) or a Robot Framework suite (.robot). The YAML file would remain the single source.
ISTQB Documentation Export
The same flow would produce a test case specification (steps, expected results, traceability through tags), regenerated from the source rather than rewritten by hand.
Local by Design
Voice mode inference is planned to run fully locally, with no cloud API: voice and screen data would stay on your machine. Without a GPU, a "deferred voice" mode is planned.
The target loop: from business gesture to replayable test
The diagram shows how CabraFlow would turn a demo into a Playwright or Robot Framework test, according to the specifications. It describes a design intention: nothing is built or measured yet.
Three planned ways in, one source file
Three ways to create a test are planned, each in a different phase of the plan. All would lead to the same versionable YAML file.
Voice Demonstration
Would open the application and speak naturally ("I select the customer profile and approve the quotation"); keyword blocks would be added as they browse. Planned for phase 4, after a scored trial.
Visual Studio
Would assemble and parameterize business micro-phrases in a visual studio: typed variables, test datasets, isolated keyword trial runs. Planned for phase 3.
Git & Text Editor
Would edit the YAML file in their own editor, with autocompletion from a JSON schema, and could read the generated Playwright code. Planned from phase 1, the first executable stage.
Planned for AI Agents (MCP): capture, replay and repair are designed to be exposed through the open Model Context Protocol (MCP), so that an agent could drive the loop like a human tester. Planned for phase 2; nothing is exposed today.
Neither a fork nor a reinvention
CabraFlow is designed as an orchestrator on top of Playwright, not a fork. It would take from Robot Framework the keyword-driven principle and composition, and from Playwright the execution engine and user-facing locators.
Flows are designed to export to Playwright or Robot Framework, so that a team does not depend on a format specific to CabraFlow.
Current R&D Program Status
As of 6 October 2026, CabraFlow is in phase 0: six specifications written on 12 August 2026 (vision, architecture, keyword language, voice mode, delivery plan, studio design). No code, no prototype, no demonstration.
We describe the actual state, in the present of the current phase: no performance figure is announced before it has been measured, and nothing will be shown before phase 1 is executable.
What exists, what will come, and on what condition
The delivery plan moves in phases, each meant to deliver something usable on its own. Only the first part of phase 0 is done.
The specifications
Six specifications written on 12 August 2026. The code repository and continuous integration remain to be set up, and the package name to be chosen.
The executable core
A command-line engine that runs a YAML flow against a small local demo application, with no selector in the tests. Exit criterion: a technical-functional tester writes a flow with autocompletion and runs it. This is the condition for any first demonstration.
MCP agents and repair
The loop exposed to agents through MCP, Playwright export (.spec.ts) and repair proposed on failure, always flagged in the report, never a silently green test.
The visual studio
A desktop application to assemble business micro-phrases as blocks, synchronized with the YAML file, with a built-in runner. The studio of this phase will not include voice mode.
Voice mode, after a scored trial
First a trial on ten recorded narrated demonstrations. Voice mode will only be integrated if over 85% of steps are correct without manual correction and perceived latency stays under 2 s. These thresholds have not been measured yet.
Ecosystem
Robot Framework export (.robot), ISTQB documentation export, continuous integration runs and shareable keyword libraries.
Project status and scope
Design the transformation of a narrated demonstration into Playwright or Robot Framework tests. Reviewed on 6 October 2026. Internal R&D status; no independent validation or customer outcome is claimed.
- Phase 0: specifications Six specifications written on 12 August 2026; no code, prototype or demonstration as of 6 October 2026. The voice mode awaits a quantified criterion (over 85% correct steps, under 2 s latency), not yet measured. Internal repository, state on 6 October 2026
Let's discuss your automated testing pipeline
Tell us about your UI automation challenges (selector fragility, business user involvement, maintenance). Your concrete cases help us test CabraFlow's design against real needs; there is no prototype to try yet.
- A technical exchange, not a sales pitch
- Your data stays in France