About hatch3r

Crack the egg. Hatch better agents.

hatch3r is an open-source CLI (and Cursor plugin) that installs a tool-agnostic agentic coding setup into any repository. It maintains a single canonical source of agent configuration — agents, skills, rules, commands, hooks, and MCP integrations — and generates native configuration for 3 AI coding platforms: Claude Code, Cursor, and GitHub Copilot. You define your setup once; hatch3r adapts it to whichever supported tool you use.

One command (npx hatch3r init) gives you the full set, optimized for your coding tool of choice. Selective init installs only what your project type and team size need.

What hatch3r is

• A generation pipeline. hatch3r owns the entire path from one canonical content source to tool-native output: .mdc rules with frontmatter scoping for Cursor, CLAUDE.md with managed blocks for Claude Code, and .github/instructions/ + .github/prompts/ + .github/agents/ for GitHub Copilot. It also emits board commands, MCP server config, and event-driven hooks.
• Tool-agnostic by design. A single source of truth with adapters for the 3 supported platforms. The canonical content ships inside the bundled npm package; adapters read from there directly.
• Audited per release. Canonical content is reviewed each release across 24 governance domains before it ships.
• Customizable without forking. Managed blocks (<!-- HATCH3R:BEGIN --> / <!-- HATCH3R:END -->) preserve your edits across updates, per-agent model selection is configurable, and .hatch3r/overrides/ is an escape hatch for user-authored canonical content.

What hatch3r is not

• Not a single instruction file. AGENTS.md is the greatest-common-denominator markdown standard for agent instructions, consumed by 20+ tools. hatch3r is complementary, not a competitor: AGENTS.md describes one file's content; hatch3r generates structured, platform-specific output across 5 artifact classes (rules, skills, commands, hooks, MCP servers) for the 3 supported platforms. Use AGENTS.md alone when one flat file suffices; use hatch3r when you need the full content-plus-tooling stack.
• Not a rule-distribution tool. hatch3r is complementary to tools like Ruler that distribute a single instruction file. hatch3r owns the full generation pipeline rather than syncing one file.
• Not a runtime sandbox. hatch3r generates configuration; it does not isolate or sandbox the AI coding tool that consumes it. Tools run with your shell credentials. See the trust model reference for what the framework does and does not guarantee.
• Not multi-tool sprawl. As of 1.9.0, hatch3r supports exactly 3 adapters (Claude Code, Cursor, GitHub Copilot). Twelve earlier adapters were removed in a deliberate scope cut to keep the supported surface current and maintainable.

Who it is for

hatch3r is for developers and teams who use AI coding tools and want a maintained, structured agentic setup instead of hand-rolling per-tool configuration. It fits:

• Solo developers who want a working setup in one command (npx hatch3r init --default).
• Teams standardizing agent behavior across Claude Code, Cursor, and Copilot from one source.
• Greenfield and brownfield projects alike — init detects your repo context (greenfield/brownfield, solo/team) and platform (GitHub, Azure DevOps, GitLab auto-detected from the git remote), and filters content accordingly.

The only prerequisite for first-run success is Node.js 22+.

The quality bar

hatch3r holds the content it generates to measurable standards rather than informal judgment. Generated end-user code is evaluated against named, checkable targets — for example:

Target	Threshold
Generated UI accessibility violations (axe-core, serious/critical)	0
Design-token adoption in generated code (color, spacing, typography)	>= 95%
Four-state surface contract coverage on generated async views (loading, empty, error, partial)	100%
Generated-service observability (OpenTelemetry) on the request path	100%
API breaking-change events on stable endpoints	0 per release
Supply-chain floor coverage	100%

These are a representative slice; the full list of content-quality thresholds ships with the framework.

Pillar framework

Every change to hatch3r serves at least one pillar on one of two axes.

Governance axis (P1-P8) — how the framework operates:

Pillar	Name
P1	CLI UI/UX Excellence
P2	Scientific & Practical Quality
P3	Adapter & External Tool Currency
P4	Comprehensive Lean Coverage
P5	Governance Self-Quality
P6	Security & Trust Governance
P7	Speed & Token Efficiency
P8	Clarification & Fan-out Discipline

Content-quality axis (CQ1-CQ9) — what the framework produces in end-user code. Each pillar is owned by a specialist agent invoked at quality gates with measurable thresholds:

Pillar	Name
CQ1	UI
CQ2	UX
CQ3	Security
CQ4	Reliability
CQ5	Testability
CQ6	Scalability
CQ7	Performance
CQ8	Maintainability
CQ9	Enhancability

See Quality-Vector Specialists for the per-pillar specialist agents and their thresholds.

Principles

• Single source of truth. One canonical content set generates every adapter's output. No per-tool copy drifts on its own.
• Lean coverage. Every artifact earns its place: no duplication, no bloat. Overlapping content is merged, not multiplied.
• Clarification before action. Ambiguous requests are resolved with a question before any irreversible change — default behavior, not an exception.
• Adversarial, root-cause quality. Findings carry measurable criteria and a causal chain, not vibes. Currency (adapter docs, CLI tool releases, security advisories) is re-checked each release.
• Security and trust by default. Atomic file writes, path-traversal guards, prompt-injection defenses, and per-agent tool allowlists are built in, not bolted on.

Next steps

• Quick Start — install hatch3r in under a minute.
• What You Get — explore the agents, skills, rules, and commands included.
• Supported Tools — the 3 supported adapters and their outputs.
• Trust Model — how distributed packs are trusted and verified.