Back to directory

Lattice

A set of composable AI skills, organized into small guardrails, multi-step workflows, and guided interviews, that push a coding assistant toward design-first, standards-aware development.

Lattice

Lattice packages engineering discipline as composable skills for AI coding assistants, aiming to stop an assistant from jumping straight to code, silently making design decisions, or losing track of constraints mid-conversation. Skills are organized into three tiers: atoms are single-principle guardrails, such as clean code, architecture, domain-driven design, or secure coding; molecules are multi-step workflows, such as design, implement, refactor, fix, or review, that compose several atoms together; and refiners are guided interviews that produce project-specific standards, tailoring how the atoms behave for a given team.

A full delivery pipeline chains a requirements stage, a design-blueprint stage, an implementation stage, and review, with separate paths for refactoring and bug fixes, and an architecture-review step that interviews an existing codebase before any of this runs. Each stage reads and writes to a living project folder that accumulates a project's standards and decisions across feature cycles. It installs as a Claude Code plugin, which Cursor can also read directly.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.