RunTheRepo v0.2.0

相比 v0.1.0

• 新增双语本地 Web Console，可自动识别公开 GitHub 仓库的最新提交、Compose 文件和应用端口，并允许手动确认或覆盖。
• 新增基于真实运行证据的阶段进度、失败原因、停止与沿用原参数重新运行，以及 HTML/JSON 报告入口。
• 新增 Web 模型设置：支持兼容 OpenAI 的 Base URL、API Key、模型发现或手动填写；密钥保存在本机 XDG 配置目录中的 owner-only 设置文件，设置接口不回显明文。
• 环境检查现在覆盖必需的 SBX 服务与诊断，Web 启动流程可拉起必备服务，并给出具体未通过原因。
• 改进 GitHub/WSL 网络可用性：加入 Windows 系统代理引导、精确提交缓存和缓存失败后的可靠回退，减少重复下载与残留目录导致的失败。
• 清理未引用的演示资产和无用运行状态，统一双语 README、使用文档与 0.2.0 包版本。

已知边界

• 结果仅适用于实际执行的 Journey 和工作负载，不构成对仓库全局安全或全局最小权限的认证。
• 当前支持的 Docker Sandboxes v0.42.0 不提供已验证的 PID 硬上限。
• Web Console 暂不提供持久化任务历史或真正的断点续跑；“继续”会沿用冻结参数重新运行。

v0.1.0-Preview

RunTheRepo 0.1.0 Preview

Run GitHub Docker Compose apps in disposable sandboxes. Verify workflows,
test hardening, and get evidence-backed reports.

Included

• Exact-commit GitHub intake and isolated Compose trials.
• Deterministic HTTP journeys, including bounded bearer reuse for declared workflows.
• Hardening experiments with verified keep-or-rollback decisions.
• JSON/HTML reports and cumulative hardened output when provenance is valid.
• Loopback-only local Web console and fail-closed sandbox cleanup.
• Simplified README with a workflow diagram and an actual report screenshot.

Verification

• Umami: operator-authored authenticated create/read/delete workflow, all six steps passing at baseline and after two kept hardening changes.
• changedetection.io: GET / baseline and two kept hardening changes passed.
• Both real runs produced reports, completed sandbox/template cleanup and left official inventory empty at production commit 7e30606. The release adds documentation only after that commit.
• Local full suite: 2238 passed, 12 skipped; coverage 85.31%. Ruff, formatting, mypy and pre-commit passed.

See validation details for exact commits, run IDs and historical failures.

Installation and limitations

This is a limited source preview, not a stable or universally compatible release.
Use the locked source installation on a pre-configured Ubuntu 24.04 WSL2/ext4 host with systemd, nested KVM and Docker Sandboxes v0.42.0.
Follow the README and setup guide; require doctor READY before a trial.

• Fresh-OS prerequisite installation is not validated.
• Real-target browser journeys are unsupported; accepted LLM contribution remains unproven.
• PID hard-bound/fork-bomb protection is not provided by the supported runtime.
• No host Docker fallback; a failed cleanup is never treated as success.
• Results apply only to the tested workflows, not global application safety.
• No public resume or durable Web history; use disposable target test credentials only.

The project has not selected an open-source license; this release does not add one.