DEV Community

Javier Leandro Arancibia
Javier Leandro Arancibia

Posted on

Same agent, different language: mago fixed a crashing test in my Zig CLI

#ai #zig #testing #automation

In a previous post I let mago — an autonomous agent team that runs over a GitHub repo — ship a feature to my Go deploy CLI. The obvious next question: does it generalize past Go, and past "add a feature"? So I pointed it at a Zig project and a crashing test.

The bug

tau is my agent-first AI CLI, reimplemented in Zig. Its test suite had one test that didn't just fail — it crashed:

error: 'agents_md.test.scanAgentsMd returns empty for bad path' terminated with signal ABRT
Build Summary: 1/3 steps succeeded (1 failed); 132/133 tests passed (1 crashed)
Enter fullscreen mode Exit fullscreen mode

132 green, 1 hard abort — the kind of thing that sits red in CI because "it's just a test."

What mago did

It read the test and found the root cause, not the symptom. The test was passing undefined as the io: std.Io argument:

// before
const result = scanAgentsMd(undefined, std.testing.allocator, std.testing.allocator, "/nonexistent") catch ...
Enter fullscreen mode Exit fullscreen mode

scanAgentsMd uses io, so in a debug build Zig traps the undefined access and aborts — before the test ever reaches the bad-path logic it meant to exercise. The fix is one line: pass a real io.

// after
const result = scanAgentsMd(std.testing.io, std.testing.allocator, std.testing.allocator, "/nonexistent") catch return;
Enter fullscreen mode Exit fullscreen mode

It opened a PR with exactly that.

Verified 

$ zig build test
Build Summary: 3/3 steps succeeded; 133/133 tests passed
Enter fullscreen mode Exit fullscreen mode

Red → green. Merged.

Two repos, two stacks, one loop

  • Last time: a feature in Go (a "did you mean" command suggester for a 32-command CLI).
  • This time: a bug fix in Zig (a crashing test, root-caused and fixed).

Same loop both times: file a GitHub issue, the agent implements it on your own LLM key, runs the repo's own build/tests, and opens a PR — verified, not blindly auto-merged. It isn't tied to a language or a framework; it works against whatever your repo's build/test command says is green.

It's good at well-scoped, verifiable work — features, fixes, tests, refactors, DX. Not "build me a startup." Honest framing: a reliable autonomous dev shop you steer with a roadmap.

Try it

CLI-only, BYOK (your Claude Code or tau key — it never resells completions), €20/mo. First 10 founding operators free during the beta, with a direct line to me:

👉 https://mago.intrane.fr

curl -fsSL https://mago.intrane.fr/install.sh | sh
mago register
Enter fullscreen mode Exit fullscreen mode

If you point it at one of your own repos — any stack — I'd love to hear how it goes.

Top comments (0)