agent-kit

Per-IDE setup walkthroughs for using @mongez/agent-kit with Claude Code, Cursor, Codex, Kiro, Gemini CLI, GitHub Copilot, Aider, and Antigravity. Each walkthrough shows the exact files agent-kit creates, the `--target` flag to pass, the reload quirks, and a copy-pasteable `package.json` snippet.

Exact commands, flags, and typical wiring for the `agent-kit` CLI (`init` / `sync` / `watch`) and its programmatic counterparts.

The `agentKit` config block in `package.json` for `@mongez/agent-kit` — every field and how it resolves: `targets` (default skill-sync agents), `pick` (allowlist) and `omit` (denylist) for filtering noisy dependency skills, and `monorepo.projects` for aggregating sibling projects. Covers pick-then-omit ordering, per-project config in a monorepo, and the root `omit` global veto.

What agent-kit is, what it does, and when an agent should reach for it — the front-door mental model covering `AGENTS.md` derivation, npm-package skill distribution, custom `--path` scan roots (monorepos / linked dev packages), nested skill folders, flat folder naming (`<pkg-slug>-<skill-path>`), and the `.agent-kit-managed` sentinel.

Symptom → cause → fix for the common @mongez/agent-kit problems: `agent-kit: command not found`, unscoped `npx agent-kit` fetching the wrong package, skills synced but the agent doesn't see them, `agentKit.pick matched no installed packages`, two-package slug collision, `sync` errors on missing AGENTS.md, a published package's `skills/` invisible to consumers, `--target` typos, `watch` not re-firing on `node_modules` edits, and a hand-authored skill folder vanishing after `sync --override`.

Using `@mongez/agent-kit` in a monorepo / full-stack repo where one session at the root must see skills from sibling projects (backend, frontend, …). Covers `--projects` / `agentKit.monorepo.projects` aggregation, the difference from `--path`, per-project `node_modules` + per-project `agentKit` config, project-dir-prefixed authored skills, shared-dependency dedupe, the root `omit` global veto, per-workspace `AGENTS.md`, and watch mode across projects.

Cross-feature wiring patterns for `@mongez/agent-kit` — per-workspace `AGENTS.md` in a monorepo, `pick`/`omit` filtering when a dependency ships skills you don't want, a CI guardrail that fails if the derived files drifted, the programmatic API for non-CLI pipelines, and watch mode for active development.

How to author and lay out skills for agent-kit — both your project's own skills in a single nested `skills/` folder at the project root, and reusable skills shipped from an npm package. Covers folder layout, nested category organization, `SKILL.md` frontmatter (`description`, `name`), the `files` field, flat destination naming, and the front-door / subskill convention.