Skip to content

AGENTS.md

Latest commit

 

History

History
51 lines (40 loc) · 3.25 KB

AGENTS.md

File metadata and controls

51 lines (40 loc) · 3.25 KB

Instructions

  • Use the @docs/ directory as the source of truth. You should list the files in the docs directory before starting any task, and update the documents as required. The @docs/ directory should always be up-to-date.
  • After completing each task, update the relevant documentation in @docs/ to reflect any changes made. In particular, @docs/design.md must accurately reflect the current state of the codebase (entities, architecture, workflows, etc.).
  • Write all comments in English.
  • Prefer enum types over strings when all variants are known at the moment of writing the code.
  • If you modified Rust code, run cargo test from the root directory before finishing your task.
  • If you modified frontend code, run pnpm test from the frontend directory before finishing your task.
  • Commit your work as frequent as possible using git. Do NOT use --no-verify flag.
  • Do not guess; rather search for the web.
  • Debug by logging. You should write enough logging code.
  • Prioritize Connect RPC-based communication for business flows over Tauri-specific bindings.

Project Structure

This is a monorepo with three main components:

Component Path Description
Main Server apps/main-server/ Go backend, task management, RPC server
Worker Server apps/worker-server/ Go worker, runs AI agents in Docker sandboxes
Desktop App apps/tauri-app/ Tauri + React frontend

Shared Rust crates live in crates/:

  • entities — Core entity definitions
  • coding_agents — AI agent abstraction, Docker sandboxing, task execution
  • task_store — Task persistence (SQLite, PostgreSQL, in-memory)
  • rpc_protocol — Connect RPC protocol definitions (Protobuf)
  • git_ops — Git operations, worktree management
  • auth — JWT authentication & RBAC
  • secrets — Cross-platform keychain access
  • worker_impl — Local worker implementation
  • config — Configuration management
  • plan_parser — YAML plan parsing

API Reference

The RPC API is defined in crates/rpc_protocol/proto/dexdex.proto. When you need to find or modify an RPC method or message type, read this file first — it lists all method names and message types, eliminating the need to guess and grep.

Types and Schemas

  • Never use any type in TypeScript. Use concrete types instead.
  • Prefer schema-based parsers (e.g., Zod) over raw JSON.parse.
  • Avoid loose interfaces and undocumented data structures. If AI has to make assumptions about the shape of data, those assumptions may be wrong and all subsequent reasoning can spiral out of control.

Utility Functions

  • Input sanitization utilities are in crates/entities/src/lib.rs (e.g., sanitize_user_input(), validate_prompt()).
  • Frontend shared utilities are in apps/tauri-app/src/lib/.
  • Read the relevant utility file before writing new helper functions to avoid duplicating existing code.
  • For Tauri commands, define a single function per command. Do NOT define two separate functions with #[cfg(desktop)] and #[cfg(not(desktop))]. Instead, use inline #[cfg(desktop)] blocks within the function body for desktop-only logic, and #[cfg(not(desktop))] to suppress unused variables and return a "not supported" error. See workspace.rs and repository.rs for reference.