CodePanda SNGroupCodePanda, your panda companion for learning to code and debug. Installation Launch VS Code Quick Open ( Ctrl+P), paste the following command, and press enter.Copied to clipboard |
CodePandaCodePanda is a classroom coding assistant with an embedded DeepSeek Harness agent loop and Cline editor integration. Sign in with your assigned student ID and username to start a task. The chat composer provides Add File, model selection, and Send. Choose DeepSeek V4.1 Flash or Kimi K3.1 Pro from the compact model menu beside Send. InstallationIn VS Code Extensions, search for For offline installation, use the delivered The extension connects to the managed classroom service automatically. No server address or API key is required. Model selection is saved between sessions. Each student receives independent daily allowances of 5,000,000 Kimi tokens and 20,000,000 DeepSeek tokens, counting input and output together. Allowances reset at 00:00 (UTC+08:00); each model’s remaining allowance appears in the expanded model menu. Both models use a 256K (262,144-token) classroom context limit, including the context meter and automatic compaction. Switch models after the active turn finishes. Server settings saved by older versions are ignored. The companion Go/Gin server must be running. Its administrator maintains the roster and model catalog. Supplier API keys are never entered into this extension. Session tokens are kept in VS Code SecretStorage. A disabled, removed, renamed, expired, or logged-out student must authenticate again. CodingCodePanda is a playful panda learning companion. Explain what your code should do, predict a result, and investigate one clue at a time. It asks one focused question at a time during debugging, offers more concrete help when you are stuck, and answers concept questions directly. It can explain or help apply a local repair; it guides course exercises without completing the whole assignment for the student. Use file editing, terminal tools, and individual approval controls from the CodePanda sidebar. Main tasks, follow-ups, commit-message generation, and compaction share the classroom model gateway. Workspace reads run automatically; operations require approval. Built-in file reads and code searches inside the opened workspace run automatically. Reading outside it, including through a symbolic link, still requires approval. Every file change, shell command, web request, MCP call or unknown tool waits for Approve, Save, or Run Command, or Reject. Approval applies only to that invocation. Old saved settings cannot grant broader automatic permissions. Questions and plain answers remain conversational. Commands execute in background processes, with output shown in the conversation. They do not open a VS Code terminal or switch focus. Cancel stops the active command. The agent loop runs inside the extension host; no separate DSH CLI, agent subprocess or sandbox is started. The interface follows VS Code's font scaling, with a 12px body-text minimum and 11px compact-text minimum. Approval buttons stay on one row, including in narrow sidebars, and touch pointers retain larger targets. Text sent during an active task steers the next model step; messages with file or image attachments remain queued for the next full turn. The account view shows the authenticated student, approval policy, and session expiry. It also provides logout and refresh controls. Service addresses and supplier keys remain hidden. CodePanda cloud sign-in, subscriptions, and provider-key configuration are not required. CodePanda stores its data separately from Cline, under Identity BoundaryThis classroom deployment intentionally uses student ID plus name without a password or second factor. A person who knows both values can impersonate the student. Administrators should distribute access only within the intended classroom and maintain the roster promptly. AttributionEditor integration is based on official Cline v4.1.18, commit |