mongez-agent-kit-recipes

Recipes

Configurations that come up when more than one tool, more than one package, or more than one environment is in play. For straightforward setup (npx agent-kit init + per-IDE --target), see Agent integrations. For flag-level CLI reference, see CLI usage.

Per-workspace AGENTS.md in a monorepo

Aggregating skills from sibling projects into one root session has its own dedicated page — see Monorepos for --projects / agentKit.monorepo.projects. This recipe is specifically about per-workspace instruction files (CLAUDE.md).

A monorepo where each package has its own conventions. agent-kit operates on the directory you point it at via --cwd, deriving one set of per-tool files per project root. To get per-workspace CLAUDE.md files, run sync inside each workspace.

repo/
├── AGENTS.md                  # repo-wide conventions
├── CLAUDE.md                  # derived
├── packages/
│   ├── api/
│   │   ├── AGENTS.md          # api-specific conventions
│   │   └── CLAUDE.md          # derived from api/AGENTS.md
│   └── web/
│       ├── AGENTS.md
│       └── CLAUDE.md

Drive it from a workspace-aware script (yarn workspaces foreach, pnpm -r, turbo run, etc.) so every package re-derives on install:

npx agent-kit sync --cwd packages/api
npx agent-kit sync --cwd packages/web

Each editor opens the closest *.md to the file being edited, so contributors working in packages/api/ get the api-specific config without manual selection.

Filter skills from a noisy dependency

By default, agent-kit scans every installed package under node_modules/ for a skills/ folder and exports them all. When a dependency ships skills you don't want — too many, off-topic, duplicate with your own — use pick (allowlist) or omit (denylist) in package.json.

{
  "agentKit": {
    "omit": {
      "@some-vendor/sdk": true,
      "@warlock.js/core": ["add-connector", "send-response"]
    }
  }
}