品質担保の為にテストするのは 変わらない 静的解析はcodacyのnormal以下の実装指摘の多くは人がやってもスルーしてるし（リファクタリングフェーズで見る） criticalやHigeの指摘は「誰が実装」しょうと全件判断（誤検おおい）するし あまり変わらない

code health with @codescene, and security with @codacy

3、/ultrareview 是 Senior Engineer 的替身 CodeRabbit 实测 recall 10%，precision 几乎不掉。 上一个做到这组合的叫 Google Tricorder，做了十年。 Qodo 在 Rakuten-SWE 任务解决率 3×。 ☠️ 砍掉：CodeRabbit/Codacy、Snyk/Checkmarx、外包 code review 服务

Today, Codacy is launching 𝗔𝗜 𝗜𝗻𝘃𝗲𝗻𝘁𝗼𝗿𝘆 - the easiest way for organizations to see what 𝗔𝗜 𝘁𝗵𝗲𝗶𝗿 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀 𝗮𝗿𝗲 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘂𝘀𝗶𝗻𝗴, and act on it. Try it →codacy.com/ai-inventory Every day brings a new tool, a new feature, a new way to build software with AI. Teams are facing relentless waves of adoption pressure, with little time to adapt - or risk being left behind. In big wave surfing, they call it a hold-down: being forced underwater by massive waves without being able to come up for air. That’s the reality for engineering and security leaders right now. 𝟵𝟭% 𝗼𝗳 𝗔𝗜 𝘁𝗼𝗼𝗹𝘀 𝗶𝗻 𝗼𝗿𝗴𝗮𝗻𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗿𝗲 𝘂𝗻𝗺𝗮𝗻𝗮𝗴𝗲𝗱. Only 19% of companies have visibility into what AI tools their developers actually use. And with the EU AI Act’s high-risk obligations coming into force in August, the urgency of visibility is increasing. Codacy’s AI Inventory gives organizations a 𝘄𝗮𝘆 𝘁𝗼 𝘁𝗮𝗸𝗲 𝘀𝘁𝗼𝗰𝗸 𝗼𝗳 𝘁𝗵𝗲 𝗔𝗜 𝗰𝗼𝗱𝗶𝗻𝗴 𝘁𝗼𝗼𝗹𝘀, 𝗺𝗼𝗱𝗲𝗹𝘀, 𝗠𝗖𝗣 𝘀𝗲𝗿𝘃𝗲𝗿𝘀, 𝗹𝗶𝗯𝗿𝗮𝗿𝗶𝗲𝘀, 𝗮𝗻𝗱 𝗔𝗣𝗜 𝗸𝗲𝘆𝘀 𝘂𝘀𝗲𝗱 𝗮𝗰𝗿𝗼𝘀𝘀 𝗲𝘃𝗲𝗿𝘆 𝗿𝗲𝗽𝗼 - without surveys or manual cataloging. You connect your repos, and Codacy builds the inventory from what’s actually committed: config files, dependencies, commit metadata, and more. We already support more than 28 tools and models, with no setup required beyond connecting your repos on Codacy. Over the past year, I’ve spoken with many engineering and security leaders who have 𝗻𝗼 𝗿𝗲𝗮𝗹 𝘄𝗮𝘆 𝘁𝗼 𝗸𝗻𝗼𝘄 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝗶𝗿 𝘁𝗲𝗮𝗺𝘀 𝗮𝗿𝗲 𝘂𝘀𝗶𝗻𝗴. Even among organizations with an AI policy, very few are enforcing it. This was, by far, the most requested feature we heard: give us a way to see what’s actually happening, and let us act on it. If you’re already on Codacy, 𝘆𝗼𝘂𝗿 𝗶𝗻𝘃𝗲𝗻𝘁𝗼𝗿𝘆 𝗶𝘀 𝗯𝗲𝗶𝗻𝗴 𝗯𝘂𝗶𝗹𝘁. If you’re not, connect your repos and see what your developers are actually using.

Stop wasting time on manual code reviews Here are 5 AI tools that actually improve your code in 2026 @GitHubCopilot @DeepCodeAI @codacy @swimm_io @AskCodi

Most Used DevOps Tools by Category: 🔹 Scripts: Perl, Python, Bash 🔹 CI/CD Tools: Jenkins, GitLab, CircleCI, Bamboo, TeamCity 🔹 Build Tools: Gradle, Maven, Apache Ant, BuildMaster 🔹 Report, Statistics & Analytics: Tableau, Nagios, Allure, SolarWinds, Dynatrace 🔹 Static Code Analysis Tools: SonarQube, Codacy, ReSharper, Coverity 🔹 Test Automation: SoapUI, JMeter, Postman, Katalon, Cypress, Selenium, UFT One 🔹 Deploy Tools: Ansible, Chef, Puppet, CakePHP 🔹 Virtualization & Containerization: Hyper-V, Amazon ECS, Mesos, Docker, Kubernetes 🔹 Source Code Management Tools: GitHub, Bitbucket, Perforce, Visual Studio

Yesterday @AnthropicAI announced @claudeai, which hunts for bugs and reviews the code Claude itself produced. I can't be the only one not spotting the conflict of interests here. You can't be the player AND the referee at the same time. This is one of the reasons why I've made a small investment into @codacy, a tool to ensure security and code quality for AI-accelerated coding. I believe the current era and the future will need more human control and guardrails, not less, and for that... independent tools will be very much needed. Thanks @jaimefjorge for allowing me in! 🤝

How it works: install a skill, type /resolve-agent-reviews on any PR, and walk away. The agent: 1. Triages each comment (true issue, false positive, or unsure) 2. Fixes real issues, dismisses false positives with an explanation 3. Asks you when it's unsure 4. Replies to every thread, auto-resolving if wanted 5. Polls for new findings and repeats until bots go quiet Three skills let you target what matters: all reviews, bot-only, or human-only. Under the hood, the CLI is a token-efficient, purpose-built alternative to gh for review comments (inspired by agent-browser). Agents can list, filter, reply, resolve, and watch with a single tool call instead of fragile multi-step gh api chains. Also works great from your terminal. Supports Claude Reviews, Codex Reviews, Copilot, CodeRabbit, Sourcery, Codacy, SonarCloud. Works with any harness that supports Agent Skills (Claude Code, Cursor, Codex, etc). Github: github.com/pbakaus/agent-rev…

Anthropic created the flood and is now selling the levee. Claude Code grew engineer code output 200% year over year. More code means more PRs. More PRs means more review bottleneck. More bottleneck means more demand for Code Review at $15-25 per pop. This is a self-reinforcing revenue loop built into the product architecture. The better Claude Code gets at writing code, the more Code Review revenue scales. No other AI company has this dynamic. GitHub Copilot bundles review into the subscription. CodeRabbit competes on price. Anthropic is the only one charging premium per-review and benefiting when their own coding tool increases volume. The internal numbers tell the real story. Before Code Review: 16% of PRs got substantive comments. After: 54%. On large PRs over 1,000 lines, 84% surface findings averaging 7.5 issues. Engineers disagree with less than 1% of flags. That last number is the one to watch. Less than 1% false positive disagreement means Anthropic solved the noise problem that killed every previous automated review tool. The reason developers ignored Codacy, SonarQube, and legacy linters wasn’t capability. It was crying wolf. When 30% of flags are wrong, engineers stop reading them. When 99% are right, engineers start depending on them. And they’re explicitly not competing on speed. 20 minutes per review. Multiple agents in parallel, cross-verifying findings, ranking severity. They’re burning serious compute on purpose because the buyer isn’t the developer. The buyer is the VP of Engineering who just watched a one-line auth change almost break production. Anthropic called it insurance, not productivity. That’s a category-creation move. Every other AI review tool is selling “review PRs faster.” Anthropic is selling “stop shipping bugs to production.” Different buyer, different budget, different price sensitivity. At $2.5B Claude Code run-rate revenue, even a small attach rate on Code Review across enterprise is a massive new line. And it compounds every time Claude Code gets better at writing code. The pricing discourse is a distraction. The strategy is a flywheel.

But with a specific tool or just the agent? Snyk is just short of free runs so other like tools like codacy and so on.

English is the S-tier programming language now. Dana Lawson (Netlify CTO) on Agent Experience replacing Developer Experience, the death of “LGTM,” and why flaky tests are still ruining everyone’s day. Thanks @jaimefjorge and @Codacy for the convo: youtube.com/watch?v=-q3k3E5F…

SonarCloud charged by lines of code, and was a nightmare to roll out at scale. So LSports switched to Codacy. Two years later: test coverage up from 7% to 70%, and zero new critical security issues. Worth a read if you want to move away from SonarQube👇 blog.codacy.com/how-lsports-…

⚙️ VS Code started as a 10-person team building an editor. Ten years later, it’s used by millions of developers, supports 90K extensions, and has become a foundation many AI tools build on. In this behind-the-scenes conversation with @codacy, we talk about the early pivots between web and desktop, the bet on open source, why performance and core editing still matter as much as AI, and how agents are starting to shape what comes next. 📖 Read the full interview: blog.codacy.com/how-vs-code-…

We will be there. Bring your toughest questions for Dana. 🔥🤙

Counting down. Dana’s going to bring the heat. 🔥 See you Feb 5!

Yes. Dana Lawson is joining @codacy next Thursday! If you have tough questions about AI, infra, and how we ship faster without extra ops overhead, drop them here and tune in. 👇

Here's the repo and installation instructions: /plugin marketplace add jaimefjorge/codacy-plugin /plugin install codacy /codacy:setup Then run with /codacy:security-review github.com/jaimefjorge/codac…

I'm awake! 7am, Let's Gooooooo!