2.9.30
Released 8 Nov 2026
CycleMantle 2.9.30 is a Windows 10/11 agent that reads your PHP code, runs your test command and shows every result in the chat, so you decide which new tests to keep.
src/Cart.phploadedfunction total1 matchtests/CartTest.phpsavedyour test commandexit 0your coverage commandreport savedcoverage.txtloadedReleased 8 Nov 2026
Setup program for Windows
64-bit desktop app
Languages of the app interface
No sign-up needed to run
Auto-run, Ask first, Plan only
Unit testing means checking one small piece of code, such as a function, against a known input and an expected result. Most projects start strong. Then the untested corners pile up, and writing the missing tests by hand takes more patience than people have. CycleMantle reads your PHP files, drafts the tests as diffs and runs your test command in one of 3 approval modes.
You open the project folder and ask for tests of a class named Cart. The agent uses read_file to load the source, grep to find the public functions, and write_file to save tests/CartTest.php. Ask first stops there. You read the diff, accept it, and Shell runs your test command. If a test should expect an exception, ask for an assertion message for exceptions, so a failing run says which error was expected. Reject any test you do not understand. The result stays 1 file in your folder.
read_file src/Cart.php → loaded
grep function total → 1 match
write_file tests/CartTest.php → saved
Shell your test command → exit 0Shell runs your coverage command, which writes a report into 1 folder, and read_file opens that report so the agent can list the functions whose lines never ran. It then proposes a test for each one as a diff, written with write_file. Teams that track automation test coverage metrics often chase one big number. Numbers can mislead. The report only says which lines ran, so treat the figure as a to-do list, accept a test, run the command again, and read the new report before you move on.
Shell your coverage command → report saved
read_file coverage.txt → loaded
write_file tests/CartTest.php → savedShell runs the command you normally type in cmd.exe or PowerShell and reports the real exit code, so a failing test shows its output in the expanded card and read_file can open the test file it names. Read the output. Start in Plan only if you are unsure: the agent writes a plan and runs nothing until you switch the mode. Then pick Ask first, and every change waits for your OK. Run the command again after each accepted edit, 1 edit at a time.
To mock a REST API from an OpenAPI file, ask the agent to read the file with read_file and to write a JSON file of fixed replies with write_file for the calls your code makes. The test then reads that JSON instead of contacting a server. Keep it local. Use this only for services you own or describe in your own files, then open the new fixture in the diff, compare 1 reply with the OpenAPI text, and run your tests once more.
A chat assistant answers. CycleMantle 2.9.30 acts, and every step becomes a card with its real result, so you check evidence instead of trusting a description of what might have happened.
| Typical chat assistant | CycleMantle | |
|---|---|---|
| Running tests | You copy the command and run it yourself | Shell runs your test command in the chat |
| Result | A description of what should happen | The real exit code and the full output |
| Files | Code in a reply that you paste by hand | Test files written as a diff you approve |
| Approval | Nothing to approve or reject | Ask first pauses before changes; Plan only runs nothing |
PHP developers who inherited code without tests are the first group, and they need a safety net before they change anything. Start small. The agent reads one class, drafts the first 3 tests as a diff, and you decide which ones stay in the repository.
Developers who already use test coverage tools are the second group. Small teams benefit most. The agent reads the report, names the files with untested lines and writes the missing tests for you to review, so nobody copies line numbers by hand.
Students learning to write tests are the third group, and Plan only suits them best. Nothing runs. You read the plan, learn why each check exists, and then let the agent write the 4 files for the first exercise while you follow each card.
Unit testing is the practice of checking one small piece of code, such as a function or a class method, in isolation with a known input and an expected result. It proves one case. A passing test does not show that the whole program is right.
Yes. This is the official CycleMantle website, and the download button gives the original build of the app, whose SHA-256 checksum you should compare with the published manifest before you run it. Questions go through the Send feedback button inside the app, and the app is independent of the PHP project.
A coverage report lists which lines or branches of your code ran while the tests ran. It points to untested code. It cannot tell you that the tests check the right things, so read each new test the agent proposes before you accept it.
The simplest way is to give the agent your test command in the chat. Shell runs it in the environment you chose, and the card shows the real exit code and output. Read the card. Use Ask first if the agent may also change files, or Plan only to see a plan first, and press Accept only for edits you understand.
No. The agent can misread code or write a weak test. Review every diff, run your test command yourself when in doubt, and keep your version control history so you can step back. Passing tests and a high coverage figure are not a guarantee.
Yes, with a cloud model: your prompt and the file text the agent reads go to that provider. With a local model nothing leaves your PC. Chats, working folders and API keys stay on your PC, and the app has no telemetry, so the only text that travels is what you send to the endpoint you picked. You choose it.
CycleMantle runs on Windows 10 and 11, 64-bit, and the setup program is about 34 MB. No account is needed. Code coverage tools and test runners are the ones already installed on your PC, and the agent only runs the commands you give it.
Download the setup program, compare the checksum, connect a model and then open one PHP project to ask for tests of a single class. Review every diff. Version 2.9.30 is current.
Download CycleMantle for Windows ↗Every page of this website is available in 20 languages, while the app interface itself has 13.
What the agent can do for your tests ↗ Connect a cloud, local or built-in model ↗ Step-by-step guide to made-up test data ↗ Download the Windows setup program ↗ Answers about testing and the app ↗ What this website and the app record ↗