Before anyone fixes a bug, someone has to see it. A maintainer adds needs-repro to a bug report; an agent follows the steps on the default branch, writes the smallest test that fails because of the bug, and posts on the issue the exact command, its output and the test itself. It ends with reproduced, or cannot-reproduce and precise questions for the author. It fixes nothing.
What this factory does
needs-repro is removed in both cases, so adding it again (after the author answered) starts a new attempt.
Why the test is not committed
The reproducer cannot open a pull request. A pull request that only adds a failing test would fail your CI, and a factory with checks refuses to publish code that fails them. So the test lives in the comment, in full, with the path where it belongs: whoever fixes the bug, a person or an agent, starts from it, and it becomes the regression test of the fix.
Bugs in the interface
When the bug shows in the app, the reproducer follows the browser-evidence skill: it starts the app with the dev command and walks the steps with agent-browser. It does not push, so it cannot attach screenshots: it describes what each step showed (the address, the element, the exact text on screen).
Tree
openfabriq/
├── factory.yaml setup, and the test and dev commands
├── agents/
│ └── reproducer/agent.md follows the steps, writes a failing test, fixes nothing
├── automations/
│ └── reproduce-bugs/ when: issue.labeled needs-repro
└── skills/
└── browser-evidence/SKILL.md walk a scenario in a real browser, describe what it showedLeast privilege
The reproducer has read, shell and commands, and no edit: it writes its test with the shell in its own worktree, runs it, and deletes it before finishing. It may only comment, add its two labels and remove needs-repro. It never opens a pull request, so nothing it writes reaches a branch.
Make it yours
- Copy the
openfabriq/folder to the root of your repository. - In
factory.yaml, replacerepository,maintainers,setupand thetestanddevcommands. - For bugs in the interface, install agent-browser on the engine's machine:
openfabriq doctorprints the command, from the skill'scompatibility(the engine's Docker image has it). - Run
openfabriq plan: it lists the labels it will create (needs-repro,reproduced,cannot-reproduce). openfabriq apply, thenopenfabriq start.- Add
needs-reproto a bug report.
With 02-sdlc, a reproduced bug goes on with a person's label: needs-investigation when the cause is unknown, ready-for-agent when the fix is obvious.



