The built-in demo runs the real reducer on a small, trusted fixture. This preview shows what it removes and what the result still does. It is a maintainer-created synthetic example, separate from the real-project case studies.
On macOS/Linux, the demo creates three source files:
source/
reproduce.py
input.txt
unused.txt
input.txt starts with three lines. reproduce.py emits
REPOMIN_DEMO_FAILURE and exits with code 1 only when a complete line is
NEEDLE; without it, the script emits a different message and exits 2.
The reducer must preserve both the target message and exit code.
-remove-before
NEEDLE
-remove-afterThe result keeps reproduce.py and an input.txt containing only NEEDLE
and its newline. It deletes unused.txt entirely. The source directory is
retained alongside the result, so you can compare them.
With Python 3.9+ and uv 0.13.4 or newer, run this exact pre-alpha development version in a new workspace. Keep uv up to date as described in the Python security notice. These commands use Bash/Zsh on macOS/Linux:
uvx --from "https://github.com/fly1d/repomin/releases/download/v0.1.0.dev15/repomin-0.1.0.dev15-py3-none-any.whl" \
repomin demo ./repomin-demoThis does not install into the project's Python environment. The first
invocation downloads the tool; the fixture and reduction need no network,
project dependencies, or API key. An existing workspace is never overwritten.
Choose another directory if repomin-demo already exists.
The macOS/Linux result has this layout:
repomin-demo/
source/ original three files
reduced/
reproduce.py unchanged failure script
input.txt NEEDLE plus a newline
reduced.repomin/
report.json recorded reduction evidence
REPOMIN.md human-readable reduction receipt
The demo checks the exported payload's exact fingerprint automatically.
The observed macOS/Linux output for v0.1.0.dev14 includes:
Reduced: 3 files / 855 bytes -> 2 files / 276 bytes in 14 attempts.
Removed: unused.txt and two unrelated input lines.
Kept: reproduce.py and input.txt (only NEEDLE remains).
Validated: exact payload fingerprint.
On Windows, the fixture uses reproduce.ps1, so file names and byte counts
differ. See the PowerShell walkthrough.
A small fixture demonstrates the mechanics. A useful bug reproduction also needs a failure signal that rejects setup errors and unusable results. Start with the real-repository quick start, or reply in the public pilot thread with a licensed repository and exact revision, reproduction command, and target failure signal. No installation or completed report is required for that initial fit check.
For an inspectable real-project result, browse the
public tsdown fixture
and its case study. The validated payload had eight
files; its public presentation adds a README and .gitignore. It was a
maintainer-run feasibility trial, not evidence of independent adoption.
ReproMin is pre-alpha. Its default host backend executes the supplied command with your account; only use repositories and commands you trust. A payload fingerprint verifies bytes, while replay executes a command to check the failure in fresh copies. Neither proves a globally minimal result or diagnoses the bug.