Imported from ramarivera/design-palace (
AGENTS.md). Install upstream withnpx skills add ramarivera/design-palace. Copyright stays with the author.
AGENTS.md — design-palace
Node project. Follow global agent rules from ~/.grok/AGENTS.md (and walked AGENTS.md family). This file supplements with project-local conventions.
Session-Start Protocol (Mandatory)
Before any substantive work in this directory:
- Check only this root (non-recursive) for:
./CLAUDE.md,./AGENTS.md,./openspec/,./.beads/. - Read all found instruction files.
- Load
verification-before-completionskill (treat as standing gate before any completion claim). - Load
yeet,projects-ledger-protocol,think-aheadetc. as relevant from global skill set. - Run memory recall via available MCP context tools (nia__context etc.) + toolbox ledger check for "design-palace".
- Acknowledge loaded files.
- Only then proceed.
If any required files were missing at start of session, they must have been created with starter structure first.
Key Project Rules
- Task runner is mise, not pnpm. All project tasks, extra tools, env, and commands go through
mise.toml.- Use
mise run <task>(ormise <task>). pnpm run/ barepnpmscripts are for compatibility only or when explicitly calling a package bin.- New tasks, tool pins, env vars, and aliases belong in
mise.toml.
- Use
- Pin Node and pnpm versions in
mise.toml(never rely on global or corepack alone for reproducible env). - Prefer
mise install+mise run installfor bootstrap. - After changing mise.toml, run
mise trust -y && mise installas needed. - Use Nushell-compatible commands when providing instructions to Ramiro. Fallback bash/zsh only when nu can't express it (label clearly).
Commands (use mise)
# First time / after pulling
mise trust -y && mise install
# Install deps (defined in mise.toml)
mise run install
# List available tasks
mise tasks
# Typical workflow
mise run dev
mise run build
mise run test
mise run lint
mise run verify # composite if defined
See mise.toml for current tasks and tool versions.
Git & Hygiene
- Never bypass pre-commit or use destructive git (
reset,checkout --,stashunless last resort,force-push) without explicit approval. - When shipping: prefer
yeetprotocol if invoked; otherwise normal careful commit + push that preserves other agents' work. - Commit messages: conventional (feat/fix/chore/docs/etc).
Adding Code / Features
- Requirements <99% clear? Ask.
- Once scope agreed, do not silently shrink or phase it.
- Before claiming done: re-run verification-before-completion gate (re-read latest agreed scope, run the proving commands fresh, list missing items if any).
- For any project-idea-like signal in chat, consider
projects-ledger-protocolcapture in toolbox.
Project Notes
- Name: design-palace
- Stack: Node + pnpm + mise (expand as real code is added)
- This is currently a minimal initialized skeleton.
Update this file as the project gains real structure, scripts, architecture decisions, or local overrides.
Global cross-refs (always loaded first):
~/.grok/AGENTS.md- Relevant skills (verification-before-completion, yeet, toolbox-rules, etc.)