Lattice Init

Lattice Init

Required Skills

Read apply skills order:

1. framework:knowledge-priming -- Load project context understand what project is what already exists

Workflow

Step 1: Scan the Project

Detect signals about project understand shape existing Lattice state.

Language/framework detection -- check files in repo root:

• package.json → Node.js / TypeScript
• tsconfig.json → TypeScript (confirm over JavaScript)
• go.mod → Go
• pom.xml or build.gradle or build.gradle.kts → Java or Kotlin
• Cargo.toml → Rust
• requirements.txt or pyproject.toml or setup.py → Python
• Gemfile → Ruby
• *.csproj or *.sln → C# / .NET
• Package.swift → Swift

<!-- synced with language-idioms-refiner "Detect the language" -- edit both -->

If multiple language markers are found in the repo root, note all of them and ask the user which is the primary stack before continuing.

Directory structure -- list top-level dirs. Identify common patterns:

• src/, lib/, app/ → source code
• test/, tests/, spec/ → test suites
• docs/ → documentation
• cmd/, internal/, pkg/ → Go project structure
• domain/, infrastructure/, application/ → layered architecture

Existing .lattice/ state -- check what Lattice artifacts already exist:

• .lattice/config.yaml → central config (check for language key)
• .lattice/standards/language-idioms.md → language idioms refiner output
• .lattice/standards/knowledge-base.md → knowledge priming output
• .lattice/standards/architecture.md → architecture refiner output (clean architecture, hexagonal, modular monolith, or custom style)
• .lattice/standards/clean-code.md → clean code refiner output
• .lattice/standards/ddd-principles.md → DDD refiner output
• .lattice/standards/review-standards.md → review refiner output
• .lattice/context/ → feature context documents (count them)
• .lattice/learnings/operational-learnings.md → accumulated operational learnings (managed by learning-harvest atom)
• .lattice/reviews/review-log.md → review log
• .lattice/requirements/index.md → check shape: if epic sections and feature tables are written directly inside it (no epics/ directory alongside) and requirements_layout is absent from config, flag as legacy layout — upgrade available
• .lattice/verification.yaml → verification stages config consumed by the verification runner
• .lattice/scripts/run-verification.sh → vendored verification runner

Step 2: Present Findings

Present:

## Project Scan Results

**Project**: [detected language/framework] at [repo root]
**Structure**: [key directories found]

### Lattice Setup Status

Running mode: **[customized -- standards docs active below / built-in defaults -- full functionality]**

- `.lattice/config.yaml`: [exists / not created yet]
- Language: [detected language / language key from config / not detected]
- Language idioms: [.lattice/standards/language-idioms.md / built-in default]
- Knowledge base: [.lattice/standards/knowledge-base.md / built-in default]
- Architecture standards: [.lattice/standards/architecture.md / built-in default]
- Clean code standards: [.lattice/standards/clean-code.md / built-in default]
- DDD standards: [.lattice/standards/ddd-principles.md / built-in default]
- Review standards: [.lattice/standards/review-standards.md / built-in default]
- Context documents: [N found / none]
- Review learnings: [found at .lattice/learnings/operational-learnings.md / none]
- Review log: [found at .lattice/reviews/review-log.md / none]
- Requirements layout: [sharded / legacy — upgrade available / not found]
- Verification suite: [.lattice/verification.yaml configured / not set up]

STOP (fresh install): if no .lattice/ state exists at all AND no legacy requirements layout was detected — create the minimal .lattice/config.yaml shown in Step 3, tell the user: "Lattice is ready. It runs on built-in defaults with full functionality. Refiner interviews that pin your team's conventions are optional — ask for them anytime." Skip to Step 4. Do not present the customization menu unprompted.

STOP: If .lattice/config.yaml and all core standards docs exist AND no legacy requirements layout was detected: Tell user "Lattice fully configured." Skip to Step 4.

STOP: if a legacy requirements layout was detected, do not skip on that basis alone — present it as a gap in Step 3 even when everything else is fully configured.

Step 3: Guided Setup

Reached only when something needs attention (a gap above) or the user asked to customize. Frame every item below as optional refinement -- Lattice is fully functional without any of it.

Priority order:

1. Requirements layout upgrade (if legacy layout detected in Step 1) -- "Your requirements index uses an older layout that hand-edits one shared file per feature, which causes merge conflicts when multiple developers work in parallel. The current layout shards it by epic and generates rollups from feature files instead of hand-editing them. One-time migration; does not touch any feature file's content beyond a link repointing (a rare exception is surfaced and confirmed, never silent). This upgrade only matters if your team keeps requirements in this repo -- teams tracking requirements in an external system (Jira, Linear, etc.) can skip it."

2. Knowledge-priming-refiner (if .lattice/standards/knowledge-base.md missing) -- "Captures project identity -- tech stack, architecture, directory layout, conventions. Every other skill uses this context make better decisions."

3. Language-idioms-refiner (if .lattice/standards/language-idioms.md missing) -- "Defines how your language expresses engineering patterns -- error handling, type system, naming, testing, DI. Multiple atoms use this to adapt pseudocode defaults to your language. Fast interview: proposes language-idiomatic defaults, you confirm or adjust."

4. Architecture-refiner (if .lattice/standards/architecture.md missing AND project has source code dir) -- "Defines project architecture standards — layer structure, dependency rules, validation checklist. Supports multiple styles: clean architecture (default), hexagonal / ports & adapters, modular monolith, or custom."

5. DDD-refiner (if .lattice/standards/ddd-principles.md missing AND project has domain folder or domain-like structure) -- "Captures aggregate design rules, entity patterns, domain event conventions so DDD atom enforces domain modeling style."

6. Clean-code-refiner (if .lattice/standards/clean-code.md missing) -- "Tailors coding standards -- function size limits, complexity thresholds, naming conventions. Defaults work well most projects, so optional."

7. Review-refiner (if .lattice/standards/review-standards.md missing) -- "Customizes how review molecule works -- atom loading rules, severity levels, report format, scope rules. Defaults work well most projects, so optional."

8. Verification setup (offer whenever .lattice/verification.yaml or .lattice/scripts/run-verification.sh is missing) -- Lattice's independent done-gate, deliberately not wired into any skill: .lattice/verification.yaml lists this project's real checks (build, unit, integration), a vendored runner reads that config and runs stages deterministically, and a verifier subagent returns one light verdict -- sessions stay quiet on green, hear only genuine failures on red. This is an additional capability, not part of any workflow: present what it does, answer questions about it, and let the user decide -- many projects will not want automated gating.

   If the user accepts:

   1. Detect candidate commands from the detected stack (e.g. package.json scripts, go test ./..., pytest, cargo test, dotnet test). Propose a stage list and confirm. Detect nothing runnable → say so and set up nothing: an empty stage list can only ever produce error results.

   2. Write .lattice/verification.yaml: version: 1, optional runsDir, and one stage per confirmed command (name + command). Verification always stops on the first failed stage. Fold working-directory changes or timeouts into the command itself when needed. Schema reference: the header comment in run-verification.sh.

   3. Vendor the runner to .lattice/scripts/run-verification.sh -- copy it from the Lattice distribution. Locate the source relative to how Lattice is installed, never guess a path: Claude Code plugin → $CLAUDE_PLUGIN_ROOT/scripts/run-verification.sh; Codex plugin → scripts/run-verification.sh two levels up from this skill's own directory (Codex plugins ship scripts as a sibling of skills/, not via an env var — there is no Codex plugin-root variable); framework dev checkout (neither applies) → scripts/run-verification.sh at the repo root. Exactly one of the two artifacts already existing → complete only the missing one.

   4. Wire the gate: explain that nothing runs until something invokes it, then offer to append this marked block to the project's instruction file. Target file: CLAUDE.md if running as Claude Code, AGENTS.md if running as Codex or another AGENTS.md-convention host. If both files already exist in the project, append to both — either host may read the project later. If neither exists yet, create the one matching the current host only.

      <!-- lattice:verification -->
      Before declaring any work done, run this project's verification suite (.lattice/verification.yaml): spawn the `verifier` subagent when the host supports subagents; otherwise run `.lattice/scripts/run-verification.sh .lattice/verification.yaml` and read summary.json from the printed run directory. Green → reply in one line. Red → headline, failed stage name(s), their log paths; never open or paste a log into the session. Never mark work complete while any stage fails.
      <!-- /lattice:verification -->

      On confirmation, append verbatim between the markers -- touch nothing else in the file; skip silently if the markers already exist. On decline, print the block and say where to paste it later.