Imported from debuglebowski/slayzone (AGENTS.md). Install upstream with npx skills add debuglebowski/slayzone. Copyright stays with the author.

SlayZone

You and I are developing SlayZone, a desktop task management app with integrated AI coding assistants (Claude Code, Codex, Gemini, and more).

You (this instance) is actually running inside SlayZone now. We are dogfooding SlayZone, i.e. using SlayZone to develop SlayZone.

You are able to interact with the running SlayZone application via the CLI. But you MUST load the slay skill before running any slay CLI command. Do not guess subcommands or flags — the skill has the full reference.

You can omit the task-id unless you want to target another task — commands auto-resolve to your current task: $SLAYZONE_TASK_ID is used if set, otherwise the task bound to $SLAYZONE_SESSION_ID (always set in a task terminal) is looked up. Trust the resolution: just run the command, don't check or echo the env vars, and pass an explicit task-id only when you deliberately target a different task.

Communication Style

Default to caveman ultra mode for entire session. Load the caveman skill at session start. Skill's own boundaries (code/commits/PRs stay normal) still apply.

Engineering Mindset

Assume near-infinite dev capability. Always pick most sustainable, robust long-term solution — never the quick hack. But never drop existing functionality to get there. Migrate, refactor, preserve behavior.

Stack

• Runtime: Electron 41
• Frontend: React 19, TailwindCSS 4, Radix UI
• Database: SQLite (better-sqlite3)
• Backend: Convex (cloud), Express (local API)
• Terminal: node-pty, xterm.js
• AI: Claude Code, Codex, Gemini, Cursor, OpenCode, Copilot, Qwen + custom modes
• Protocols: MCP (Model Context Protocol)

Architecture

See ARCHITECTURE.md for system architecture and PHILOSOPHY.md for structural principles.

Commands

Commit Messages

Conventional Commits:

Scope optional: feat(terminal): ...

E2E Testing Rules

• TDD: Always run tests FIRST to see them fail, then fix code. Never write tests alongside code and assume they pass.
• useRef + useEffect for DOM measurement: If a component has early returns (loading/null guards) before the measured element, useEffect([], []) runs when the ref is still null. Use a callback ref instead.
• Hook lifecycle across tabs: Hooks' useEffect only runs on mount. Tabs stay mounted with display: none — seeding settings and navigating doesn't re-trigger effects. Test by opening a NEW task (fresh hook mount).