Example Projects
Real, runnable examples that ship in the Qontinui repositories
Every example below lives in source control and runs out of the box. The demo workflows are seeded into Qontinui Runner on first launch (or created by a setup script), and the configuration samples demonstrate execution modes for the Python library. Use them to learn the verification-agentic loop and the model-based automation patterns hands-on.
The Core Idea
In each demo you define what needs to be true (verification checks — tests, linters, type checkers), and the AI figures out how to make it true. The runner loops the verification and agentic phases until all checks pass.
Runner Demo Workflows
Three end-to-end demos in the runner repository. Each has a setup phase that creates its workspace, a verification phase that runs a test suite, and an agentic phase where the AI fixes or implements the code.
Fix Buggy Calculator
Simple~30sDemonstrates: The verification-agentic loop on a focused bug fix.
- A Python calculator ships with 3 arithmetic bugs (subtract adds, divide multiplies, modulo does integer division)
- Verification runs the test suite and reports failures
- The AI reads the failing tests, identifies the root cause, and edits the implementation
- Verification re-runs — all tests pass
Implement from Tests (TDD)
Medium~35sDemonstrates: Building code from a test specification — true test-driven development.
- The test file is the specification; the implementation starts empty
- Verification fails initially because the functions don't exist
- The AI reads the tests and implements five string utilities — reverse_words, title_case, count_vowels, is_palindrome, truncate
- Verification passes (22 tests) in a single iteration
Fix Data Pipeline
Complex~40sDemonstrates: Validating computed output against expected values across a multi-step pipeline.
- A Python pipeline reads a sales CSV, computes revenue, and writes a JSON report
- Validation checks the output against manually calculated expected values
- The AI fixes a type-conversion bug (string→float), an arithmetic bug (multiply→divide), and an aggregation bug (min→max)
- Validation confirms the corrected report
Running them: The demos are seeded automatically on the runner's first launch. If they're missing, run the create_demo_workflows.ps1 script in the examples/demo-workflows directory. They require Python and the Claude CLI to be installed.
Configuration Samples
The runner repository also ships standalone configuration files that demonstrate the library's execution modes and patterns. Each is a complete JSON config you can load directly.
Mock Mode Example
config-mock-mode.jsonRuns a click / wait / type workflow with executionMode set to mock — waits resolve instantly and no real GUI is touched.
Real Mode Example
config-real-mode.jsonThe same patterns executed against the live GUI through the hardware abstraction layer.
Screenshot Mode
config-screenshot-mode.jsonMatches against captured screenshots rather than the live display for reproducible scene testing.
Recursive Verification
config-recursive-verify.jsonDemonstrates the verification loop that re-checks state after each action until conditions hold.
Task Continuation
config-task-continuation.jsonShows how a multi-step task carries context forward across continuation points.
Notepad Automation
notepad_automation.jsonA classic visual-automation example driving the Windows Notepad application.
Python Library
To use Qontinui programmatically, explore the core Python library source. It includes the mock subsystem, the find/action engine, and the state-management model that power every example above.
qontinui (core library)↗
Python source for the visual automation engine — model, mock, actions, and state management. See the in-repo Python examples and module READMEs.