Releases: eatdrop/patchproof
Release list
PatchProof v0.1.2 — Real Bottle Integration
中文
v0.1.2 为 PatchProof 增加首个真正独立的第三方开源缺陷验证案例:固定 Bottle 修复前提交与准确的上游源码修复 Hunk,通过独立复现测试、359 项 Bottle 原生回归测试和不同的边界测试完成四阶段验证。
新增
- Bottle HTTP 条件请求优先级真实集成案例。
- 独立
bottle-integrationDocker CI 与内容寻址审计产物。 - 包式
unittest discover的安全 top-level 支持。 .tpl文本快照支持。- 自动忽略测试目录中的标准
__pycache__生成物。
验证证据
- 80 项 PatchProof 自动化测试通过。
- Bottle:补丁前复现失败、补丁后通过、359 项上游回归通过、1 项独立边界测试通过。
- 目标 Bottle checkout 前后快照一致。
- 合并后 GitHub Actions 四项检查全部通过。
English
v0.1.2 adds PatchProof's first genuinely independent third-party open-source defect case. It pins Bottle's pre-fix commit and exact upstream source-fix hunk, then establishes the four-phase result with an independently authored reproduction, 359 Bottle-native regressions, and a distinct edge case.
Added
- A real Bottle HTTP-precondition precedence integration case.
- A dedicated Docker-backed
bottle-integrationCI job with content-addressed audit artifacts. - Safe unittest package discovery top-level support.
.tpltext-snapshot support.- Standard
__pycache__artifacts are ignored in external test bundles.
Evidence
- 80 PatchProof automated tests passed.
- Bottle: fail before patch, pass after patch, 359 upstream regressions passed, and 1 independent edge case passed.
- The target Bottle checkout retained the same before/after snapshot.
- All four post-merge GitHub Actions checks passed.
PR: #3
Post-merge CI: https://github.com/eatdrop/patchproof/actions/runs/30799759460
Wheel SHA-256: 86269161839fbe74350d81eb1483fcf19f139fa1823a9f316844a76d86af3280
PatchProof v0.1.1
PatchProof v0.1.1 — Clearer Value, Faster Proof
这个版本优化的不只是 README,而是首次使用体验:用户现在可以在三分钟内运行真实四阶段 Demo,并独立重验验证回执的内部完整性。
This release improves more than the README—it improves first use. A user can now run a real four-phase demo in under three minutes and independently re-verify a validation receipt's internal integrity.
新增 / Added
make demo-local:无需 Docker 和第三方依赖的一键可信本地 Demo。make demo-local: a one-command trusted-local demo with no Docker or third-party dependencies.patchproof verify-receipt:重新验证严格结构、阶段语义和内容哈希。patchproof verify-receipt: revalidate strict structure, phase semantics, and content hashes.- 基于真实夹具输出的 1200×700 README 终端演示图。
- A 1200×700 README terminal visual based on real fixture output.
- Demo 重复执行回归测试,总测试数提升至 71。
- A demo repeatability regression test, bringing the suite to 71 tests.
门面与定位 / Front Door and Positioning
- 首屏明确“是什么、给谁、解决什么”。
- The first screen states what it is, who it serves, and what problem it solves.
- 新增“为什么普通单次 CI 不够”的逐项对比。
- Added a direct comparison explaining why typical one-pass CI is insufficient.
- 新增三分钟 Quick Start、适用/不适用范围、当前成熟度和双语 FAQ。
- Added a three-minute quick start, fit/non-fit guidance, current maturity, and bilingual FAQ.
- 明确当前是面向 Coding Agent 基础设施与评测工程师的 Alpha 协议实现,不声称通用测试平台或生产多租户沙箱。
- Clarified that this is an alpha protocol implementation for coding-agent infrastructure and evaluation engineers, not a universal test platform or production multi-tenant sandbox.
验证 / Verification
- 71 项自动化测试通过 / 71 automated tests pass.
- Python 3.11 与 3.12 CI 通过 / Python 3.11 and 3.12 CI pass.
- GitHub Actions Docker Shadow Validation 通过 / GitHub Actions Docker shadow validation passes.
- Wheel 完成构建、隔离安装与版本导入验证 / The wheel is build-, isolated-install-, and version-import-verified.
- GitHub 实际 README 渲染已验证:演示图、表格和折叠区正常 / The rendered GitHub README was verified: image, tables, and disclosure section load correctly.
PatchProof v0.1.0
PatchProof v0.1.0
首个公开 Alpha 版本:把 AI 生成补丁从“看起来合理”转化为可重复、可审计的四阶段验证协议。
First public alpha release: turn AI-generated patches from “looks plausible” into a repeatable, auditable four-phase validation protocol.
核心能力 / Highlights
- 仓库快照绑定的 Proposal Hash 与精确完整哈希审批。
- Repository-snapshot-bound proposal hashes and exact full-hash approval.
- 在一次性副本中应用 Unified Diff,真实仓库保持不变。
- Apply unified diffs in disposable copies while preserving the real repository.
- 补丁前失败复现、补丁后复现、完整回归、外置隐藏测试。
- Fail-before reproduction, pass-after reproduction, full regression, and external hidden tests.
- 固定镜像、断网、只读、非 root 和资源受限的 Docker Runner。
- Digest-pinned, network-disabled, read-only, non-root, resource-bounded Docker runner.
- 内容寻址 JSON 回执和中英双语 Markdown 报告。
- Content-addressed JSON receipts and bilingual Markdown reports.
- 两个可独立复用的 Codex Skills:
verifiable-agent-audit与agent-eval-builder。 - Two independently reusable Codex Skills:
verifiable-agent-auditandagent-eval-builder.
验证 / Verification
- 70 项自动化测试通过 / 70 automated tests pass.
- Python 3.11 与 3.12 CI 通过 / Python 3.11 and 3.12 CI pass.
- GitHub Actions 中真实 Docker 四阶段 Shadow Validation 通过 / Real Docker four-phase shadow validation passes in GitHub Actions.
- Wheel 已完成构建、隔离安装与版本导入验证 / The wheel is build-, isolated-install-, and version-import-verified.
边界 / Scope
v0.1 仅支持有界 UTF-8 文本、Unified Diff 和 Python unittest。Docker 加固不是生产级恶意多租户沙箱,审批人身份是声明式元数据,哈希回执提供完整性而不是签名者真实性。
v0.1 supports bounded UTF-8 text, unified diffs, and Python unittest only. Docker hardening is not a production hostile multi-tenant sandbox; approver identity is declared metadata, and hash receipts provide integrity rather than signer authenticity.