command-creator

# Command Creator $ARGUMENTS Create a new Claude Code slash command following ai-toolkit conventions. ## Workflow 1. **Capture intent** -- what should the command do and who invokes it? 2. **Pick scope** -- project command, user command, or plugin command? 3. **Define frontmatter** -- description, argument hint, allowed tools, model if needed 4. **Write command body** -- instructions for Claude, not explanation for the user 5. **Add validation context** -- expected output, constraints, follow-up checks 6. **Test and iterate** -- invoke the command with representative arguments ## Command Locations - Project: `.claude/commands/<name>.md` - User: `~/.claude/commands/<name>.md` - Plugin: `plugin-name/commands/<name>.md` In `ai-toolkit`, command-development guidance belongs in skill form and any reusable command templates should live under `app/skills/<skill-name>/templates/` or KB docs. ## Frontmatter Template ```yaml --- description: "One-line help text shown in /help" argument-hint: "[target]" allowed-tools: Read, Grep, Bash model: sonnet --- ``` ## Command Authoring Rules - Write commands as **instructions for Claude**, not marketing copy for the user - Keep the first paragraph action-oriented and deterministic - Use `$ARGUMENTS` early when the command takes user input - Prefer numbered phases for multi-step workflows - List required checks explicitly (`lint`, `tests`, `docs`, etc.) - Avoid hidden assumptions about project structure unless the command is project-sp...