loam 0.0.2

Codebase intelligence and anti-AI-slop for Dart & Flutter: dead code, cycles, duplication, complexity and architecture drift, behind a baseline/ratchet CI gate.

Codebase intelligence & anti-AI-slop for Dart & Flutter.

---

loam.dev is the product; loam is this CLI command and pub.dev package. Web: getloam.dev · Source: github.com/silvio-l/loam

loam catches the structural drift and AI-generated slop that dart analyze never sees — behind a baseline/ratchet CI gate that never paints a grown codebase red on day one. Built on the Dart analyzer package: semantically accurate, project-wide, offline by default.

What it catches

Structural drift (deterministic, semantic)	AI-slop (deterministic + optional LLM)
Unused public exports, files, members	Empty / swallowing catch blocks
Circular dependencies	Narrative filler comments
Code duplication (AST-normalised)	Ungrounded // ignore:
Complexity hotspots + health score	Duplicated helpers, dead guards
Architecture-boundary violations	Hallucinated / superfluous abstractions

What makes it different

• 🌱 Semantic, not regex — resolved Dart element model + project-wide graphs.
• 🔒 Baseline / ratchet gate (default) — freeze today's findings; only new ones fail CI.
• ♻️ Reproducible even with an LLM — verdicts cached by sha(code)+prompt@ver, fixed thresholds decide. Same code = cache hit = stable verdict, zero token cost.
• 📄 Self-contained HTML report — one offline file; toggle findings, copy a fix-prompt for your AI agent.

Install

dart pub global activate loam

Quick start

🚧 Early development. The walking skeleton wires the pipeline and the tracer rule (unused-public-exports); the remaining capabilities land as individual rules. Commands below are the target surface.

loam scan                 # full audit: every active rule, whole repo
loam baseline --write      # freeze the accepted state
loam gate                  # CI: ratchet — only new findings fail

Status

Walking skeleton in progress. Founding spec, domain glossary and architecture decisions live in the repository.

License

MIT © 2026 Silvio Lindstedt