# Create Skills And Rules > Create or update Claude Code skills and Cursor rules/skills with proper format and frontmatter, and configure OpenCode Use when this capability is needed. - Skill: `tomevault-io/create-skills-and-rules` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/create-skills-and-rules` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/create-skills-and-rules/raw - Safety review: pending (external: skill-scanner PASS, skillspector CAUTION) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/create-skills-and-rules --- # Skill: create-skills-and-rules Use this skill when you need to **create or update skills and rules** for Claude Code (`.claude/`), Cursor (`.cursor/`), and OpenCode (`opencode.json`). ## Directory structure ``` .claude/ rules/ *.md – always-loaded project instructions skills/ /SKILL.md – step-by-step guides, invocable or auto-matched by description .cursor/ rules/ *.mdc – Cursor rules with description/globs/alwaysApply frontmatter skills/ *.md – flat files, same content as .claude/skills/ (Cursor reads them as reference) opencode.json – OpenCode config: instructions array points to .claude/rules/*.md ``` **Cross-tool compatibility:** - **Claude Code** reads both flat files and directory-based skills from `.claude/skills/` - **OpenCode** reads **only** directory-based skills (`.claude/skills//SKILL.md`) — flat files are ignored - **Cursor** reads flat files from `.cursor/skills/*.md` as reference documents - Rules are loaded by OpenCode via `opencode.json` `instructions` array **Standard format:** Always use directory-based skills (`/SKILL.md`) for compatibility with all tools. ## Step 1 – Decide: rule or skill | Type | Purpose | When loaded | |---|---|---| | **Rule** | Constraints, conventions, best practices | Automatically, based on `globs` or `alwaysApply` | | **Skill** | Step-by-step implementation guide | When description matches task, or invoked via `/skill-name` | Rules say **what to do / not do**. Skills say **how to do it step by step**. ## Step 2 – Create a skill (`.claude/skills//SKILL.md`) ### Create the directory and file ```bash mkdir -p .claude/skills/my-skill-name # then create .claude/skills/my-skill-name/SKILL.md ``` ### Required frontmatter ```yaml --- name: my-skill-name description: Short description of what this skill does and when to use it --- ``` | Field | Required by | Description | |---|---|---| | `name` | OpenCode, Claude Code | Must match directory name exactly. Lowercase, hyphens only, regex `^[a-z0-9]+(-[a-z0-9]+)*$`, max 64 chars | | `description` | All tools | When to use — Claude/OpenCode match this to decide auto-invocation | ### Optional frontmatter fields (Claude Code only) ```yaml --- name: my-skill description: What this skill does user-invocable: true disable-model-invocation: false allowed-tools: Read, Grep, Bash context: fork agent: Explore argument-hint: [filename] model: sonnet --- ``` | Field | Default | Description | |---|---|---| | `user-invocable` | `true` | Show in `/` menu | | `disable-model-invocation` | `false` | Prevent auto-loading by Claude | | `allowed-tools` | – | Tools allowed without asking (e.g. `Read, Grep, Bash`) | | `context` | – | Set to `fork` to run in isolated subagent | | `agent` | – | Subagent type when `context: fork` (`Explore`, `Plan`, etc.) | | `argument-hint` | – | Hint for autocomplete (e.g. `[issue-number]`) | | `model` | inherited | Model override (`sonnet`, `opus`, `haiku`) | OpenCode ignores unknown frontmatter fields, so these are safe to include. ### Dynamic substitutions in skill content - `$ARGUMENTS` – all arguments passed when invoking - `$ARGUMENTS[0]`, `$1` – specific argument by index - `${CLAUDE_SESSION_ID}` – current session ID - `${CLAUDE_SKILL_DIR}` – directory containing SKILL.md ### Body structure ```markdown --- name: my-skill-name description: Build X with Y and Z --- # Skill: my-skill-name Use this skill when you build **X** using Y and Z. ## When to use - Bullet list of scenarios ## Step 1 – First step Explanation + code example. ## Step 2 – Second step Explanation + code example. ``` Keep skills under 500 lines. Use clear step numbering. Include real code examples. ## Step 3 – Create a Claude Code rule (`.claude/rules/.md`) ### Frontmatter ```yaml --- description: Short description of this rule's purpose globs: **/front/src/**/*.{vue,js,ts} --- ``` | Field | Description | |---|---| | `description` | What this rule covers (used for matching) | | `globs` | File patterns that trigger this rule (e.g. `**/*.js`, `**/services/**/*.js`) | ### Backend / LiveChange service rules — standard `globs` Rules for **server-side** LiveChange code (models, actions, triggers, service layout) should use a **comma-separated** `globs` line so they still match when a project stores services under different folder names (many teams copy `.cursor/rules` into other repos): ```yaml globs: **/services/**/*.js, **/server/**/*.js, server/**/*.js ``` | Pattern | Covers | |---|---| | `**/services/**/*.js` | Trees such as `live-change-stack/services//` | | `**/server/**/*.js` | Any nested `server/` directory (e.g. `app/server/`, `packages/foo/server/`) | | `server/**/*.js` | Backend rooted at top-level `server/` | Frontend rules keep their own globs (e.g. `**/front/src/**/*.{vue,js,ts}`). Do not drop the `server` variants for backend-only rules — otherwise Cursor will miss files after a layout change. ### Body structure ```markdown --- description: Rules for X development globs: **/*.js --- # Title ## Section 1 - Rule bullet points - Code examples ## Section 2 - More rules ``` Rules are concise. Lead with the constraint, then show a code example. ## Step 4 – Create Cursor rule (`.cursor/rules/.mdc`) Same content as the Claude rule, but with Cursor-specific frontmatter: ```yaml --- description: Short description of this rule's purpose globs: **/front/src/**/*.{vue,js,ts} alwaysApply: false --- ``` | Field | Default | Description | |---|---|---| | `description` | – | What the rule does; Cursor uses this to decide when to apply | | `globs` | – | Comma-separated file patterns (e.g. `**/*.tsx, src/**/*.js`) | | `alwaysApply` | `false` | If `true`, applies to every session regardless of context | **Notes:** - Do NOT quote glob patterns in frontmatter - Keep rules short (target 25 lines, max 50 lines for best Cursor performance) - The `.mdc` extension is required for Cursor - For **backend** LiveChange rules, use the same **`globs`** standard as in Claude rules: `**/services/**/*.js, **/server/**/*.js, server/**/*.js` ## Step 5 – Register rules in OpenCode (`opencode.json`) OpenCode reads `.claude/skills//SKILL.md` natively for skills (no extra step needed). But rules must be registered in `opencode.json`: ```json { "$schema": "https://opencode.ai/config.json", "instructions": [ ".claude/rules/live-change-backend-architecture.md", ".claude/rules/live-change-frontend-vue-primevue.md" ] } ``` When you **create a new rule**, add its path to the `instructions` array in `opencode.json`. Project rule **`live-change-node-toolchain-fnm`** should stay **first** (or early) in `instructions` so agents always load the requirement to run `node`, `npm`, `npx`, and `tsx` via `fnm exec` using `.node-version` / `.nvmrc`. When you **create a new skill**, no `opencode.json` change is needed — OpenCode discovers skills from `.claude/skills//SKILL.md` automatically. When a skill or rule shows shell examples that invoke **`node`**, **`npm`**, **`npx`**, or **`tsx`**, use the **`fnm exec -- …`** form (see `.claude/rules/live-change-node-toolchain-fnm.md` and skill `live-change-node-toolchain-fnm`). **Important:** OpenCode ignores the `globs` frontmatter from Claude Code rules. All instructions listed in `opencode.json` are always loaded. ## Step 6 – Mirror Cursor skills (`.cursor/skills/.md`) Copy the skill content to `.cursor/skills/` as a **flat file** (Cursor reads flat `.md` files as reference): ```bash cp .claude/skills/my-skill-name/SKILL.md .cursor/skills/my-skill-name.md ``` The content is the same (including `name` in frontmatter). Cursor ignores unknown frontmatter fields. ## Step 7 – Mirror to sub-projects If the project has sub-projects with their own `.claude/` and `.cursor/` directories, copy the files there too: ```bash for dir in automation auto-firma; do # Skills: directory format for .claude, flat for .cursor mkdir -p "$dir/.claude/skills/my-skill-name" cp .claude/skills/my-skill-name/SKILL.md "$dir/.claude/skills/my-skill-name/SKILL.md" cp .claude/skills/my-skill-name/SKILL.md "$dir/.cursor/skills/my-skill-name.md" # Rules cp .claude/rules/my-rule.md "$dir/.claude/rules/" cp .cursor/rules/my-rule.mdc "$dir/.cursor/rules/" done ``` ## Naming conventions - Use lowercase with hyphens: `live-change-frontend-editor-form` - Must match regex `^[a-z0-9]+(-[a-z0-9]+)*$` (OpenCode requirement) - Prefix with domain: `live-change-frontend-*`, `live-change-backend-*` - Skills describe actions: `*-editor-form`, `*-range-list`, `*-action-buttons` - Rules describe scope: `*-vue-primevue`, `*-models-and-relations` ## Checklist - [ ] Shell examples for Node/npm use `fnm exec --` per `live-change-node-toolchain-fnm` - [ ] Directory created: `.claude/skills//SKILL.md` - [ ] Frontmatter has both `name` (matching dir) and `description` - [ ] `.cursor/skills/.md` mirrored (flat file, same content) - [ ] `.claude/rules/*.md` created (if rule) - [ ] `.cursor/rules/*.mdc` created with `globs` + `alwaysApply` (if rule) - [ ] Backend rules use standard `globs`: `**/services/**/*.js, **/server/**/*.js, server/**/*.js` (when the rule targets LiveChange server code) - [ ] `opencode.json` `instructions` array updated (if new rule) - [ ] Sub-projects updated (automation, auto-firma) --- > Source: [live-change/live-change-stack](https://github.com/live-change/live-change-stack) — distributed by [TomeVault](https://tomevault.io).