Hanzo Kata · Test repair
Test. Repair. Verify.
Autonomous loop.
Kata keeps your tests green. It triages a failing CI run, finds the regression, writes the fix and runs the suite again before anything lands.
Read the docs →terminal
kata run --suite tests/unit
12 passed · 1 failed (test_token_expiry)
kata triage
Kai decision: patch · confidence: 0.96 · root cause: off-by-one jitter
kata repair --model enso-ultra
synthesized AST edit in auth/tokens.py · re-running test suite...
13 passed · 0 failed · 100% green verification
The loop
From a failing test to a verified fix.
Kata never submits speculative code: Kai triages every failure, Enso writes the fix, and your own test suite decides whether it lands.
01 · Observe
Catch the failure
Reads the stack trace, exit code and test log from pytest, cargo test, vitest or go test.02 · Decide
Choose what to do
Kai scores retry, patch, roll back or hand to a person, as probabilities, and writes no text.03 · Write
Write the fix
Enso picks the model that writes the edit, weighing quality, latency and budget.04 · Verify
Prove it
The candidate diff runs in an isolated microVM sandbox. Only a change whose whole suite passes lands.Capabilities
Built for the pipeline.
Test repair, triage and the choice of model, in the CI you already run.
Test repair, end to end
Finds regressions and flaky tests, traces each to its root cause, writes the smallest diff and checks that the suite passes.Triage by Kai
No prose to parse: Kai returns a probability for each action, patch, roll back, retry or approve.Models picked by Enso
Each attempt runs on enso-auto, enso-flash or enso-ultra, trading cost against accuracy.Runs in your pipeline
In GitHub Actions, GitLab CI or a pre-push hook in your terminal.Questions
Asked plainly.
How does Kata differ from Hanzo Dev and Gen?
Dev is the coding agent that plans and writes whole features. Gen runs any coding agent on Hanzo’s tools and models. Kata repairs failing tests in CI: it triages each failure and lands a fix only when the suite passes.
Which test frameworks does Kata support?
Kata natively supports pytest, vitest, jest, cargo test, go test, and custom test runners emitting standard TAP or JUnit XML output.
Where can I find the open-source examples and recipes?
The open-source repository is at github.com/hanzoai/kata with runnable examples in Python, TypeScript, Go, Rust, and cURL, along with a standalone Rust recipes crate.
How do I authenticate with Hanzo Cloud?
Log in once with `hanzo auth login` or export `HANZO_API_KEY` from platform.hanzo.ai/keys.
Read
The agent, the loop, the documentation.
Explore the source code, recipes, and documentation.
Paste and ship
Paste this into any agent. It reads the skill manifest and calls Kata from there.
Agent prompt
Read https://hanzo.ai/skill.md and use Hanzo Kata in my project. Start with: hanzo kata run