claude-automation-recommender

The full skill.

Original instructions from the publisher’s SKILL.md

# Claude / Cursor Automation Recommender

Analyze codebase patterns and recommend tailored automations. **Read-only** — analyze and report; do not create or modify files unless the user explicitly asks you to implement recommendations afterward.

**Source:** [claude-code-setup plugin](https://github.com/anthropics/claude-plugins-official/tree/main/plugins/claude-code-setup)

## Cursor vs Claude Code paths

| Concept | Claude Code | Cursor IDE |
|---------|-------------|------------|
| Skills | `.claude/skills/<name>/SKILL.md` | `.cursor/skills/` or `~/.cursor/skills/` |
| Slash commands | Plugin / project commands | `.cursor/commands/*.md` → `/filename` |
| Rules | `CLAUDE.md`, `.claude/` | `.cursor/rules/*.mdc`, `AGENTS.md` |
| Hooks | `.claude/settings.json` | `.cursor/hooks.json` + `.cursor/hooks/*` |
| MCP | `.mcp.json`, `claude mcp add` | Cursor Settings → MCP, or project MCP config |
| Subagents | `.claude/agents/*.md` | Task tool (`explore`, `generalPurpose`, etc.) + skills |

When the user works in **Cursor** (this environment), prefer Cursor paths in recommendations. Mention Claude Code equivalents when relevant for teams using both.

## Output guidelines

- **Recommend 1–2 per category** — top picks only, unless the user asks for one category (then 3–5).
- **Go beyond reference lists** — use web search for tools/frameworks specific to the repo.
- **End with**: user can ask for more in any category; offer to help implement.

## Automation types

| Type | Best for |
|------|----------|
| **Hooks** | Automatic actions on tool/session events (format, lint, block edits) |
| **Subagents** | Specialized parallel reviewers (security, performance, a11y) |
| **Skills** | Packaged expertise and repeatable workflows |
| **Plugins** | Bundled skills (Claude marketplace / Cursor global skills) |
| **MCP Servers** | External tools (docs, DB, browser, GitHub, etc.) |
| **Slash commands** | Quick workflows (`/test`, `/pr-review`) |

## Workflow

### Phase 1: Codebase analysis

Gather project context (adapt commands for Windows PowerShell when needed):

```powershell
# Project markers
Get-ChildItem -Name package.json, pyproject.toml, Cargo.toml, go.mod, CMakeLists.txt, AGENTS.md, CLAUDE.md -ErrorAction SilentlyContinue

# Existing AI config
Test-Path .cursor, .claude, AGENTS.md, CLAUDE.md
Get-ChildItem .cursor -ErrorAction SilentlyContinue
Get-ChildItem .claude -ErrorAction SilentlyContinue
```

Also inspect: `engine/`, `src/`, `tests/`, CI configs, formatters/linters, `.env*`, lockfiles, MCP configs.

**Key indicators**

| Category | Signals | Informs |
|----------|---------|---------|
| Language / build | CMake, Cargo, package.json, etc. | Hooks, MCP |
| Frontend | React, Vue, Slint UI | Playwright MCP, UI skills |
| Backend / APIs | Express, FastAPI, custom servers | API doc skills |
| Database | Prisma, SQL, Supabase | DB MCP |
| External SDKs | Stripe, AWS, Vulkan | context7, domain MCP |
| Testing | pytest, CTest, Playwright | Test hooks, subagents |
| CI/CD | GitHub Actions | GitHub MCP |
| Docs | OpenAPI, AGENTS.md | Convention skills |

### Phase 2: Generate recommendations

Use reference files in this skill directory:

- [references/mcp-servers.md](references/mcp-servers.md)
- [references/skills-reference.md](references/skills-reference.md)
- [references/hooks-patterns.md](references/hooks-patterns.md)
- [references/subagent-templates.md](references/subagent-templates.md)
- [references/plugins-reference.md](references/plugins-reference.md)

**Cursor hooks:** map Claude `PostToolUse` / `PreToolUse` patterns to `.cursor/hooks.json` events such as `afterFileEdit`, `preToolUse`, `beforeShellExecution`, `beforeSubmitPrompt` (see create-hook skill if needed).

**Cursor subagents:** recommend Task tool usage + project skills (e.g. `code-reviewer`, `feature-dev` in `~/.cursor/skills/`).

### Phase 3: Recommendations report

Only **1–2 per category**; skip irrelevant categories.

```markdown
## Automation Recommendations

Analyzed your codebase. Top 1–2 picks per category:

### Codebase profile
- **Type**: …
- **Framework**: …
- **Key libraries**: …
- **AI tooling detected**: Cursor / Claude Code / both

---

### MCP servers
#### [name]
**Why**: …
**Install (Claude Code)**: `claude mcp add …`
**Install (Cursor)**: Settings → MCP, or project MCP config

---

### Skills
#### [name]
**Why**: …
**Cursor**: `.cursor/skills/<name>/SKILL.md` or `~/.cursor/skills/<name>/`
**Claude Code**: `.claude/skills/<name>/SKILL.md`

---

### Hooks
#### [name]
**Why**: …
**Cursor**: `.cursor/hooks.json` — event: `afterFileEdit` / `preToolUse` / …
**Claude Code**: `.claude/settings.json`

---

### Subagents / parallel agents
#### [name]
**Why**: …
**Cursor**: Task tool + skill, or dedicated explore/review prompt
**Claude Code**: `.claude/agents/<name>.md`

---

### Slash commands
#### /[name]
**Why**: …
**Cursor**: `.cursor/commands/<name>.md`

---

**Want more?** Ask for another category.
**Want implementation help?** Ask to set up any recommendation.
```

## Decision framework

### MCP servers
External services, live docs, browser testing, GitHub/Linear/Slack, cloud infra.

### Skills
Repeated workflows, project conventions, templates/scripts, user-only side effects (`disable-model-invocation: true`).

### Hooks
Post-edit format/lint, block sensitive files, validation after edits.

### Subagents
Security/performance review, parallel audits, large codebases.

### Plugins / command bundles
Multiple related skills; team standardization (e.g. official `feature-dev`, `claude-code-setup`).

## Configuration tips

- **Team MCP**: commit shared MCP config so the team matches.
- **Cursor project hooks**: prefer `.cursor/hooks.json` in repo for shared behavior.
- **Permissions (Claude Code)**: `.claude/settings.json` allow lists for hook tools.
- **Headless Claude (CI only)**: `claude -p "…"` when Claude Code CLI is installed.