Pick the path that matches where you are.
Each path is a short orientation — a couple of paragraphs and the commands you'd run. Depth is a click away in the README.
The commands below are for Claude Code — the richest integration. Using Cursor, Windsurf, GitHub Copilot, Aider, Continue, Junie, Codex, Zed, or another coding agent? Drop one small rule file into your project. Snippets for each tool: docs/setup.md. Same workflow; only the invocation differs.
I'm evaluating.
You've heard of LID and you want to know if it's for you. LID frames itself differently from the spec-driven-development tools you may have looked at: not as an "SDD methodology" but as a structured way to write what your coding agent compiles — your design at a few levels of detail, from broad direction down to the granular claims tests verify. Your time goes into the design docs; the agent writes the specs, tests, and code from those.
The fastest loop: skim the HLD for the framing, skim the README for the mechanics, try the quickstart for the feel. Twenty minutes and you'll know.
The HLD names the problem (intent gaps in agentic coding), the reframe (LID as a structure for intent), the mechanisms (narrowing the latent space, linkage, semantic legibility, tests-first), and the adoption modes. The examples page links to two projects — one clean and minimal, one real and messy.
Not sure spec-driven development itself is worth the effort? Fair question. The anti-patterns page names when LID is the wrong choice — honestly, without overselling.
LID even ships a coach: /lid-coach reads an already-running LID project and reports back — advisory only, never editing — where your usage is drifting from the methodology's principles and where there's more value to capture. A good way to learn the discipline by seeing it applied to your own work. See the coach.
Beyond the core, an opt-in lid-experimental plugin bundles skills that have earned their place but aren't yet promoted into the core workflow — currently bidirectional-differential for auditing EARS↔code coherence. Worth a look once the basics make sense.
I'm starting a new project.
Greenfield is where LID shines hardest — nothing to reconcile, nothing to reverse-engineer. Install the plugin and tell the workflow what you want to build. It takes you through the design phases, stopping for your review each time, and only writes code once the design has settled.
/plugin marketplace add jszmajda/lid
/plugin install linked-intent-dev@jszmajda-lid
/linked-intent-dev describe what you want to build
After that, tell your agent what you want to build. You'll see an HLD first (with 2–3 architectural alternatives for big decisions), then LLDs for each component, then EARS specs, then tests-first, then code. Every phase stops for your approval.
Once a few components exist, /lid-coach reviews how well the project is applying LID's principles and points at the highest-value next steps.
I'm adding LID to an existing codebase.
Brownfield is a different shape. You have working code already; what you don't have is an articulated intent behind it. LID's /map-codebase reads every file in scope (deliberately token-intensive — accuracy requires it), proposes multiple ways to slice the system, and works backward from code to LLDs to a synthesized HLD.
/plugin install linked-intent-dev@jszmajda-lid
/plugin install arrow-maintenance@jszmajda-lid
/arrow-maintenance:map-codebase
After mapping you get an arrow — a traceable chain from intent to code — plus an audit overlay that tracks what's implemented, what has drifted, and what still needs design attention. Arrows grow as you work; /arrow-maintenance keeps them honest.
I'm trying LID on one subsystem.
Solo on a team, or piloting LID on one slice of a larger project? Scoped LID is the mode for you. The HLD is still required, but it can mark the rest of the project as "not yet specified" — you own your scope, and the discipline relaxes at the boundary.
Scoped projects use the same entry point as Full LID; the workflow asks which mode you want, or you can choose later with /update-lid.
/linked-intent-dev describe your scoped feature
# answer "Scoped" when prompted for mode
Cascade pauses at the edge of your scope, so you don't accidentally propagate incoherence into teammates' territory. As the team adopts LID or your scope expands, /update-lid supports promoting from Scoped to Full.