QualityLayer：让编码智能体按计划开发

Plan, build and review software with your coding agents. Work side by side in the App or browser, or use your favourite terminal or IDE.

QualityLayer The software factory with you in the loop Plan each change visually, then let your agents build and verify it. Work side by side in the App or browser, or use your favourite terminal or IDE. Install • Features • Docs • Website • Changelog Download the App for macOS, Windows or Linux. 7-day trial. ▶ Watch the 5:20 walkthrough Why QualityLayer “With QualityLayer, developing software with agents finally feels like doing real engineering again.” An engineer who uses QualityLayer QualityLayer keeps design decisions and verification visible while your agents work. You get How it works Your own agent and plan Claude Code, Codex, Cursor and four more run on the subscription or API key you already have. QualityLayer never charges for AI usage. A design you can review Comment on diagrams and clickable screens, then approve the Plan before code changes. Small, tested changes Your agent works in small slices, writes behavior tests before implementation and fixes failures. An independent check With independent review on, a fresh agent checks the running change against your request. Anything it cannot verify stays visible in Review. Code health checks Local checks flag complex functions and deeply nested code for your agent to review. Ask for a wider check to find duplication. Code health. Visible decisions and effort Each step records its decisions and results in your repository, with time and estimated model cost shown in the App. Choose Feature or Bugfix for a change that needs an approved Plan and recorded verification. Choose Chat for a direct conversation with Claude Code or Codex, adding QualityLayer tools when you need them. Getting Started Prerequisites A coding agent: Claude Code, Codex, Cursor, Grok, Antigravity, Pi or OpenCode, installed and signed in. Choose it in the App or browser. Other agents connect through the CLI. Your own subscription (Claude or ChatGPT plan, or API key). QualityLayer never charges for model usage. Git and a repository. Installation Download the App and open it. Setup activates your licence or starts the trial, installs the qualitylayer command and connects your agents. On a server, in WSL or in a container: curl -fsSL https://qualitylayer.dev/install.sh | bash # macOS, Linux, WSL irm https://qualitylayer.dev/install.ps1 | iex # Windows PowerShell The App updates itself; without it, run qualitylayer update. Installation guide. Uninstalling qualitylayer uninstall undoes what the installer changed; your plans stay. On Windows, even when the App is gone: & ([scriptblock]::Create((irm https://qualitylayer.dev/install.ps1))) -Uninstall Coming from Pilot Shell 11 Run pilot update to install QualityLayer, or download it directly. Keep Pilot Shell 11 alongside it until you are ready to remove it. Active Polar licences work with both products; older purchases may need help moving over. Import memories per project when you are ready. Moving from Pilot Shell. First Steps In the App or browser, choose + New, then Chat, Feature or Bugfix. Pick your repository and agent, and describe what you need. Chat opens a full-width conversation with Claude Code or Codex, without workflow steps or approval gates. Answer supported questions in the App and follow its task list above the composer when it has one. Ask it to save a plan, build in slices, make artifacts, run checks or request an independent review. In a terminal, ask: “Using QualityLayer, chat about retrying failed webhooks.” Choose Start the workflow, then Feature or Bugfix, when the work needs the full process. Plan with QualityLayer continues in Discuss; Saved plan starts at Plan from the Chat's saved plan. Both keep the same task and conversation. Approve the workflow Plan before the build. Chat guide. For a plan you already have, choose Existing plan under Feature or Bugfix. Paste it or attach a Markdown or plain-text file. QualityLayer keeps the original and opens a short execution agreement for your approval before building. For a Feature or Bugfix, describe the change: Plan retries for failed webhooks and stop after a few tries. Answer your agent's questions and comment on the Plan beside its conversation. Approve the Plan and start the build. Your agent continues through Outline, Implement and Verify. Review the result and its evidence. Checks that need you, or could not run, stay open for your decision. Where you work How to start App or browser Choose In the App to keep the agent beside the task. Your terminal or IDE Choose In my terminal to copy the request, or ask there: “Using QualityLayer, plan retries for failed webhooks.” Another agent Agents that run shell commands can use the CLI. Supported agents. To continue elsewhere, stop the active run, then ask: “Using QL, continue verification of task retry-webhooks.” Your Plan, changes and evidence stay with the task. QualityLayer runs only when you ask for it. Your first task. Skills your agent gets Setup installs six skills into each connected agent. They teach it to drive the qualitylayer command line instead of guessing: which step comes next, what needs your approval, and how to reach other agent sessions. Ask in plain language, naming QualityLayer or QL; your agent picks the skill. Skill What it does ql-workflow Plan, build, verify and review a change ql-artifact Create and revise mockups, diagrams and clickable pages ql-manage Find tasks and manage memory, sharing, settings and setup ql-peer Coordinate other agent sessions on your computer ql-quality Review code complexity, structure and duplication ql-chat Start or continue a Chat, a conversation without workflow steps Only ql-workflow starts the planning and build workflow; the others work on their own. More about the skills. The Workflow Each step you run writes a Markdown file in your repository. Existing plans start at Plan. Your agent follows a short instruction for each step and decides how to carry it out. Small changes keep a short Plan and one build step; larger work gets the detail it ne