01 / THE SIGNAL

我们发现了什么

你可以在这里查看一些示例(针对开源代码仓库):https://reports.patchwave.ai/alam00000/bentopdf/pull/913 和 https://reports.patchwave.ai/shipshapecode/shepherd/pull/355……在隐私/安全方面,我们的 SOC 2 Type II 审计正在进行中(请参阅 trust.contextbridge.ai)。

  • 来源:Hacker News(发现于 2026-10-21)
  • 证据等级:D · 发现产品或需求信号,暂未获得可核验的商业证据。
  • 商业模式:API / Usage-based
  • 主题:独立产品
  • 初筛评分:17.5/100 · 收录 1 次
#独立开发#待验证#产品发现
02 / SOURCE & EVIDENCE

证据,比故事更重要。

发现产品或需求信号,暂未获得可核验的商业证据。

规则清洗与初筛,未经人工商业核验。原文语境、实际客户和付费情况仍需自行验证。

引用与数字披露

来源类型(原作者自述/第三方测算/媒体转引)需采集端标注,本版尚未落字段。

短句引用
作者
未标注
抓取日期
来源类型
未标注
数字口径
币种
未标注
口径
未标注
披露主体
未标注
披露日期
未标注

中文辅助译文(全文)

嘿,Hacker News,我们是 Ben、Josh 和 Donnie,PatchWave 的联合创始人。PatchWave 是一款 GitHub App,它将静态分析与经过评估的智能体结合起来,用于分析 PatchWave 认为安全的 PR,并可选择自动合并这些 PR。如果某个 Dependabot PR 导致你的 CI 构建失败,修复智能体会努力让 CI 再次通过。Dependabot PR 糟透了。没人喜欢处理它们。大多数时候,它们会被置之不理,或者由资深工程师花费宝贵时间处理。在 PatchWave 之前,我经营自由职业业务,按小时向客户收费,专门做这种“维持运转”的工作。即便在我们这家小型初创公司,每逢周一早上,我们也会花时间处理 Dependabot PR。让人类参与其中、与智能体共同处理这些 PR 非常耗时,而且智能体常常会偷懒,没有真正验证上游变更对代码库的影响。随着 CVE 的数量和严重程度不断增加,快速合并安全更新会变得愈发重要。PatchWave 是一组专用智能体,它会使用静态分析工具分析我们以确定性方式收集的上下文。1. 我们获取变更日志、版本之间的代码差异、安全公告和 CI 结果。2. 智能体会在隔离的沙箱中检出你的代码,在那里使用结构化代码搜索、类型检查器以及项目自身的工具链,理解你的代码库如何使用每个正在更新的依赖项。每个 Dependabot PR 都会生成一份报告。

你可以在此处查看一些示例(针对开源软件仓库):https://reports.patchwave.ai/alam00000/bentopdf/pull/913 和 https://reports.patchwave.ai/shipshapecode/shepherd/pull/355……3. 如果 CI 失败,我们会自动派发一个修复智能体,尝试让 CI 重新变绿。对于大多数 PR,这意味着更新代码以使用更新的 API、修复代码检查/格式问题等。4. 我们有一项可选功能,当 PatchWave 认为某些 PR 安全且该 PR 符合你的自动合并策略时,可以自动合并这些 PR。我们建议的自动合并策略是,自动合并开发依赖的补丁/次要/主版本更新,以及生产依赖的补丁/次要版本更新。你可以自定义此策略,也可以完全关闭自动合并。在许多本应由人工审查的情况下,PatchWave 会跳过自动合并。例如,如果来源发生变化(比如,一个通常由 GitHub Actions 通过 Trusted Publishing 发布的包,突然改由 API Key 发布),我们会要求人工审查后再合并。Dependabot 还设置了默认启用的 3 天冷却期,以降低供应链攻击风险。如果 PatchWave 通过修复智能体进行了任何更改,我们始终会要求你在合并前进行审查。目前,PatchWave 仅支持 Dependabot 和 GitHub。我们计划在未来支持 Renovate。我们对智能体的评估主要使用 TypeScript 和 Ruby,但我们一直在扩充评估套件,以提高准确性。

修复智能体最适合处理只需要适度代码更改的 PR。对于大型更新,尤其是跨越多个主版本的更新,我们的修复智能体会放弃尝试,并为你提供一份补救计划,帮助你降低风险。在隐私/安全方面,我们的 SOC 2 Type II 审计正在进行中(见 trust.contextbridge.ai)。我们只在你选择的仓库上运行,我们的智能体每次运行都在短生命周期的 microVM 中执行,我们使用采用零数据保留的 Amazon Bedrock,并且绝不使用你的代码进行训练。我们为每个 GitHub 组织每月提供 25 个免费 PR,无需信用卡。额外 PR 的费用约为每个 1 美元,数量越多折扣越高。我们也希望帮助加速开源项目,因此请联系我们询问额外的免费额度。如果你讨厌处理 Dependabot PR,我们很希望你能试用一下并告诉我们你的看法。Josh、Donnie 和我会随时回答任何问题,并回应你提出的任何反馈。

译文由上游机器翻译生成,可能有误;判断请以英文原文为准。

英文原文(来源本站未改写)

Hey Hacker News, we're Ben, Josh, and Donnie, co-founders of PatchWave.PatchWave is a GitHub App that combines static analysis and eval'ed agents to analyze and (optionally) auto-merge PRs it deems as safe.If a Dependabot PR breaks your CI build, a fix agent works to get CI passing again.Dependabot PRs suck.Nobody likes dealing with them.Most of the time they go ignored or a senior engineer spends their valuable time dealing with them.Before PatchWave, I ran a freelance business charging clients hourly to do exactly this kind of "keep the lights on" work.Even at our small startup, we'd spend Monday mornings dealing with Dependabot PRs.

Having a human-in-the-loop working with an agent on these PRs was time consuming and, often, agents would get lazy and not actually validate how an upstream change affected the codebase.And, as the quantity and severity of CVEs increases, merging security updates quickly becomes increasingly important.PatchWave is a set of specialized agents that uses static analysis tools on context we gather deterministically. 1.We pull the changelog, the code diff between versions, security advisories, and CI results. 2.

The agent checks out your code in an isolated sandbox, where it uses structural code search, type checkers and your project's own toolchain to understand how your codebase uses each dependency being updated.Every Dependabot PR gets a report.You can see a few examples (against OSS repos) here: https://reports.patchwave.ai/alam00000/bentopdf/pull/913 and https://reports.patchwave.ai/shipshapecode/shepherd/pull/355... 3.If CI fails, we automatically dispatch a fix agent to try to get CI green again.For most PRs, this means updating your code to use a newer API, fixing linting/formatting issues, etc. 4.

We have an optional feature that allows auto-merging certain PRs when PatchWave deems them as safe and the PR aligns with your auto-merge policy.Our recommended auto-merge policy is to auto-merge patch/minor/major dev dependencies and patch/minor production dependencies.You can customize this policy or turn auto-merge off entirely.PatchWave will skip auto-merge in a bunch of cases where a human should review.For instance, if provenance changes (e.g., a package that's usually published by GitHub Actions via Trusted Publishing is suddenly published by an API Key), we require human review before merging.

Dependabot also instituted an on-by-default 3-day cooldown to reduce the risk of supply chain attacks.If PatchWave makes any changes via a fix agent, we'll always require your review before merge.Currently, PatchWave only works with Dependabot and GitHub.We plan to offer Renovate support in the future.We've eval'ed the agents the most against TypeScript and Ruby, but are always adding to our eval suite to improve accuracy.The fix agent works best with PRs that require modest code changes.For large updates, especially updates that span multiple majors, our fix agent will give up and provide you a remediation plan to help you mitigate.

On the privacy/security side of the house, our SOC 2 Type II audit is in progress (see trust.contextbridge.ai).We only run on repos you select, our agents run in a short-lived microVM per run, we use Amazon Bedrock with zero-data-retention, and never train on your code.We're offering 25 free PRs per GitHub org, per month with no credit card required.Additional PRs are billed at about $1/PR, with discounts for higher volume.We also would love to accelerate OSS projects, so reach out to ask about additional free credits.If you hate dealing with Dependabot PRs, we'd love for you to give it a shot and tell us what you think.

Josh, Donnie, and I will be around answering any questions and responding to any feedback you have.

出处https://patchwave.ai抓取日期 · 采集源 Hacker News

03 / EVIDENCE GAPS

这条还缺什么证据?

下面每条都由本条已有字段推出(等级、理由、商业模式、来源次数、是否演示), 本站不生成推测性结论;通用验证方法放在方法论页。

  • 可核验的收入或付费证据查官网定价页与付费口径;第三方数据源(如 GetLatka)只作旁证,需标注来源与时点。
  • 只有单一来源找一手站点或其他渠道是否重复出现同一产品;社区热帖数量不等于商业进展。

通用验证清单(谁有这个问题/谁愿意付费/一个人能交付哪一小步)见我们的筛选方法。

04 / SIGNAL HISTORY

发现时间线