English | Chinese

code-sidecar is a VS Code extension project that provides a lightweight sidecar workflow with a webview UI, a core task runner, and tool adapters for LLM-assisted actions.

• Focuses on a minimal, understandable architecture for learning and experimentation.
• Keeps the extension host logic in src/ and the UI in webview-ui/ for clear separation.
• Provides configurable LLM and permission settings to control behavior.

• src/: VS Code extension entry and logic (extension.ts)
• src/core/: API handler and task runner
• src/managers/: workflow helpers
• src/tools/: tool adapters
• src/ui/: sidebar and webview glue
• webview-ui/: React + Vite client, built to webview-ui/dist/
• assets/: icons for the activity bar
• dist/: build output (generated)

1. The extension activates and loads configuration.
2. Core modules handle API requests and orchestrate tasks.
3. The webview UI communicates with the extension for user-facing interactions.

• /init: Generate or update AGENTS.md at the workspace root with concise, repo-specific guidance.
• /<skill-name>: Treat the slash command name as a workspace skill hint. The agent will inspect skill folders with normal file tools, read the matching SKILL.md, then follow it before continuing.

• Supports Anthropic-compatible skill folders that contain a SKILL.md file with YAML frontmatter.
• Scans these workspace locations: .agent/skills/, .code-sidecar/skills/, and skills/.
• Skills are consumed through normal file tools by reading the relevant SKILL.md and nearby resource files when needed.

• @file path/to/file: Include the referenced file content in the prompt context.
• @file "path with spaces": Quote paths that contain spaces.
• @workspace: Include an expanded workspace structure snapshot in the prompt context.
• @workspace path/to/dir: Include a workspace tree rooted at a specific folder.
• @workspace:6 or @workspace(6): Request a custom workspace depth.
• @workspace:6 path/to/dir: Combine a depth override with a specific folder.

This extension contributes these settings:

• codeSidecar.api.baseUrl: Base URL for the LLM API endpoint.
• codeSidecar.api.model: Model name to use for LLM requests.
• codeSidecar.api.temperature: Temperature for LLM responses (0-2).
• codeSidecar.api.maxTokens: Maximum tokens for LLM responses.
• codeSidecar.permissions.allowReadByDefault: Allow file read operations without confirmation.
• codeSidecar.permissions.allowWriteByDefault: Allow file write operations without confirmation.
• codeSidecar.permissions.allowExecuteByDefault: Allow command execution without confirmation.
• codeSidecar.permissions.alwaysConfirm: Operations that always require confirmation.
• codeSidecar.maxLoopCount: Maximum number of ReAct loop iterations.
• codeSidecar.contextWindowSize: Maximum context window size in characters.

• Install dependencies: pnpm install
• Type-check, lint, and build: pnpm run compile
• Watch mode: pnpm run watch
• Lint only: pnpm run lint
• Type checks only: pnpm run check-types
• Tests: pnpm run compile-tests then pnpm test

• This project is mainly my graduation design and is intended for learning.
• The overall design and content are intentionally simple.