Core Concepts
Six terms carry most of hatch3r's mental model. Read this once before the Quick Start and the rest of the docs read faster. Each term links to the reference page that goes deeper.
Canonical source vs. adapters
hatch3r keeps one copy of your agent setup — the canonical source — bundled inside the npm package, and generates tool-specific files from it. The generators are adapters: one per supported tool (Claude Code, Cursor, GitHub Copilot). You edit intent in one place; adapters translate it into each tool's native config (.claude/ + CLAUDE.md, .cursor/, .github/copilot-instructions.md). Your repo holds only the manifest, your overrides, and the generated outputs — not the canonical content itself. Details: Content Model · Adapter System.
Agent / skill / command / rule
These are the four content types you will hear about most. Each is a markdown file with YAML frontmatter in the canonical source:
| Type | What it is | When it acts |
|---|---|---|
| Agent | A role with behavioral instructions (reviewer, implementer, security-auditor). | Invoked by a command or another agent to do one job. |
| Skill | An on-demand instruction bundle for a specific task (bug-fix, release, incident-response). | Loaded when an agent needs that procedure. |
| Command | A slash-command workflow that orchestrates one or more agents. | You run it (e.g. /feature-plan) inside your coding tool. |
| Rule | A persistent standard always in context (code-standards, security-patterns). | Applies passively to every relevant edit. |
(Two more types — hooks for event-triggered automation and checks for quality gates — round out the catalog.) See What You Get for the full inventory and the per-type reference pages.
Review loop
After an implementer agent makes a change, hatch3r runs a bounded review loop: hatch3r-reviewer inspects the diff for correctness, security, performance, and accessibility; if it raises Critical or Warning findings, hatch3r-fixer addresses them and the reviewer re-checks. The loop runs at most 3 iterations, then hands back. This is how generated code earns a quality bar without an unbounded back-and-forth. See Agentic Process.
Content profile
At init you pick a content profile that decides which artifacts get generated: Minimal (core only), Standard (recommended — core, planning, implementation, review, devops, maintenance), Full (everything), or Custom (you pick item by item). Greenfield/brownfield and solo/team context further filter the set. Change it later with hatch3r config. See the profile table.
Maturity dial
Maturity (solo · team · scaleup · enterprise) is a single dial — set once at init — that calibrates how deeply your agents invest in reliability, scale-handling, testing, and operational depth. It is not a content filter: every tier installs the same full corpus, anchored by a universal security/UI/protocol/content-quality floor that never relaxes. A higher tier means deeper investment and stricter review rigor, not a different file set. (Distinct from --team-size, which is a context filter for ctx:team-only artifacts like board management and handoffs.) See Maturity Tiers.
Next Steps
- Quick Start — install hatch3r in under a minute
- What You Get — the full catalog of agents, skills, rules, and commands