overseer: Instructions file for Codex

AGENTS.md

Instructions for AI agents working on this codebase.

Project Context

Overseer is a desktop GUI frontend for AI coding agents (Claude Code, Codex, OpenCode). See docs/OVERSEER.md for full architecture, component map, data models, and feature documentation.

Key References

• docs/OVERSEER.md — Architecture, source structure, component map, data models, implemented features, planned features
• CLAUDE.md — Development commands, pre-commit checklist, tech stack, key conventions
• docs/claude-agent-sdk.md — Claude Agent SDK reference (query API, tools, hooks, subagents, MCP, permissions, sessions)
• docs/features/ — Implemented feature specs
• docs/plans/ — Planned feature specs

Design Principles

• Private: No telemetry, no API keys, no cloud integrations. Never add analytics, tracking, or external service calls.
• Performant: No idle polling, no background refreshes. Everything is event-driven and reactive. Minimize resource usage.
• Extensible: The AgentService interface allows new agent backends without modifying existing code. Keep this boundary clean.

Development

• Package manager: pnpm
• Run pnpm checks before committing (or pnpm checks:ui for UI-only changes)
• Run pnpm test to verify changes
• Frontend: React 19 + TypeScript 5.9 + MobX + Tailwind CSS v4
• Backend: Tauri v2 (Rust) in src-tauri/

MobX Conventions

• NEVER use makeAutoObservable or makeObservable — always use @observable, @computed, and @action decorators
• Decorators go on a separate line above the member they decorate:

  @observable
  pending: PendingNote | null = null
  
  @action
  updateComment(text: string) {
    // ...
  }
  

Finding Current Models

When updating model lists in src/renderer/stores/ConfigStore.ts, check these sources for the latest available models:

• Claude: Claude models overview
• Codex: OpenAI Models or Codex Models