 General-purpose agent powered by Claude Code. Handle any task — coding, research, data analysis, writing, and automation — with strong performance on complex reasoning, long-context understanding, and nuanced architectural decisions. ## What it's good at - **Coding** — build full applications, refactor codebases, debug, and maintain complex multi-file projects - **Research** — scrape the web, synthesize information from multiple sources, produce structured reports - **Data analysis** — process CSV, JSON, and Excel files; run analysis; generate charts and dashboards - **Writing** — draft reports, proposals, documentation, and API specs with precision and nuance - **Automation** — build scripts, CLI tools, and pipelines to automate repetitive workflows - **Live preview** — run and test everything in the sandbox with a full VNC desktop ## How it works Claude Code runs in the same persistent Docker sandbox as DeepSeek Harness — full desktop, terminal, filesystem, browser, and VNC. The difference is the model: it uses Claude for reasoning, which brings strong performance on nuanced decisions, long-context understanding, and careful execution. You describe the task and Claude Code plans the approach, executes the work, and verifies the result. For large tasks it breaks the work into steps and checks each one before moving on. The sandbox is `longrunning` — it pauses when idle and resumes when you return. Everything you build persists across conversations. If Docker sandbox is not enabled, tasks execute directly on the machine where Yao Engine is installed. ## Best practices **State the outcome and the constraints.** Don't micromanage the steps — describe what you want to be true when it's done, plus any boundaries. "Summarize this 50-page PDF into a 2-page executive brief, focus on financial risks" is better than "read page 1, then page 2..." **Let it interview you for complex tasks.** For anything non-trivial, start with a brief description and ask it to ask you clarifying questions before it begins. It will surface edge cases and trade-offs you might not have considered. **Explore first, then plan, then execute.** For large tasks, separate research from action. Ask it to read and understand the situation, propose a plan, then proceed only after you approve. This avoids wasted effort solving the wrong problem. **Give it verifiable success criteria.** If you can describe what "correct" looks like — specific numbers, a format, a comparison against a reference — include that in your prompt. It will verify its own output before delivering. **Keep the conversation going.** Each message builds on everything before it. Reference documents already in the workspace, ask follow-up questions, or refine the result. "Now turn that analysis into a slide deck" works because the full context is still there. **Examples:** > Read the attached contract PDF. Extract all obligations, deadlines, and penalty clauses into a structured table. Flag anything due in the next 30 days. > Research [company name] — financials, recent news, competitive position. Produce a one-page investment memo with a recommendation. > I have meeting notes from the past month (attached). Identify recurring themes, unresolved action items, and draft a status report for leadership. > Analyze our customer support tickets (attached CSV). Find the top 5 complaint categories, suggest process improvements, and estimate impact. ## Capabilities at a glance | Capability | Details | |------------|---------| | Model | Claude (configurable) | | Languages | Python, TypeScript, Go, Rust, and anything installable | | Web access | Browse, scrape, fetch APIs | | File I/O | Read and write any file format | | Preview | Live VNC desktop — see apps running in real time | | Sandbox | Persistent Docker desktop, full terminal and browser | | Lifecycle | Long-running — pauses on idle, resumes on return | | Workspace | Permanently stored — code and files persist indefinitely | ## What's next - [DeepSeek Harness](/docs/en-us/built-in-agents/deepseek-harness) — same sandbox capabilities, powered by DeepSeek - [Web Builder](/docs/en-us/built-in-agents/website) — for full website development with deployment - [AI Experts: Find the Right One](/docs/en-us/getting-started/ai-experts-intro) — see all built-in experts
Claude Code
feedback
反馈