一个用于运行、测试和加固 GitHub Docker Compose 应用程序的 AI 代理。
Run GitHub Docker Compose apps in disposable sandboxes. Verify workflows, test hardening, and get evidence-backed reports.
Quick start · Usage guide · Setup
中文 README · English
https://github.com/user-attachments/assets/90804069-60ac-4cc4-b164-53350d152066
https://github.com/user-attachments/assets/86a19206-d66b-4400-9620-831186d2e9df
| Capability | Result |
|---|---|
| Commit-pinned trials | Know exactly which source version was tested |
| Deterministic HTTP checks | Check response status and content, including declared authenticated workflows |
| Tested hardening | Keep or roll back each change based on replayed checks |
| Usable outputs | JSON/HTML reports, evidence references and a cumulative hardened overlay when provenance is valid |
| Local console | Submit one trial, follow its status and open its report |
Before you begin: use the distro's ext4 filesystem with Ubuntu 24.04 on WSL2, systemd, accessible nested KVM and authenticated Docker Sandboxes v0.42.0 with the reviewed default-deny policy. Native Windows and host Docker execution are not supported. Follow the one-time setup guide if these prerequisites are not ready.
From your checked-out project directory, with uv installed:
uv sync --locked --all-groups
uv run playwright install chromium
uv run repotrial doctorContinue only when doctor reports READY. The setup guide covers browser OS
dependencies and proxy configuration.
For model-assisted runs, open Advanced options in the Web console. Enter an
OpenAI-compatible Base URL and API key, click Get models, then choose a
returned model or enter one manually and save. The key is persisted only in an
owner-only settings file within the XDG configuration directory. By default,
this file is ~/.config/repotrial/model-settings.json; set XDG_CONFIG_HOME to
use another XDG configuration directory. The settings API does not echo the key,
it is excluded from public job payloads, and a run passes it only to the
trusted CLI subprocess. Clear model settings to remove it.
Start the console in the same WSL Bash session:
uv run repotrial serve --port 8765Open http://127.0.0.1:8765/, enter a public repository URL, its full commit SHA and internal web port. Set your model endpoint/name when using a model. The console is loopback-only and runs one trial at a time.
Example target: Umami. Replace the model endpoint and name with your provider's
values; this command does not supply the authenticated workflow shown above.
Authentication provider keys are supplied through REPOTRIAL_MODEL_API_KEY;
see the usage guide for the
complete CLI-only instructions.
uv run repotrial inspect https://github.com/umami-software/umami \
--provider docker-sbx \
--commit-sha ca661c7057984aa98ed4f7083d84dae2f65bfcb0 \
--container-port 3000 \
--compose-path docker-compose.yml \
--model-endpoint https://MODEL-ENDPOINT/v1 \
--model-name MODEL_NAMEReports are written to artifacts//report/.
For your own HTTP checks, use --journeys-file.
| Need | Guide |
|---|---|
| Install WSL/SBX or fix connectivity | Environment setup |
| Write HTTP checks, configure a model or troubleshoot | Usage guide |
暂无开放 Issues,或尚未同步最近议题。