Adapter contract

Purpose

cofferdam_core::adapter::Adapter is a small, descriptive trait that lets a language pipeline declare its own identity: which Language it produces facts for, which files it claims, and which schema namespace its corpus/graph facts live under. It exists to formalize — as a type, not just prose in a module doc — the shape cofferdam's polylingual architecture already has (TypeScript via oxc, Rust and HTML via tree-sitter), ahead of a larger refactor that would route dispatch through it.

This is a deliberately narrow slice. It adds the trait and three thin implementations; it does not change how the engine parses or dispatches files today.

The trait

pub trait Adapter {
    fn language(&self) -> Language;
    fn file_globs(&self) -> &[&str];
    fn schema_namespace(&self) -> &str;
}

• language(&self) -> Language — the Language variant (cofferdam_core::source::Language) this adapter produces facts for.
• file_globs(&self) -> &[&str] — the glob patterns / extensions this adapter claims, mirroring what Language::from_path already encodes for that language.
• schema_namespace(&self) -> &str — the schema-extension namespace this adapter's corpus/graph facts live under (e.g. "ts", "rust").

The one-way contract (documented, not enforced)

The intended relationship between adapters and checks is one-way: adapters produce facts from source; they do not see rule configuration, and they do not construct Issues directly (that stays the Check trait's job).

This iteration does not enforce that contract. There is no runtime check that stops an Adapter impl from reading rule config or building an Issue, and no test that exercises a deliberate violation. The trait is a type-level description of intent, not a sandbox.

What implements it today

• TypeScriptAdapter (crates/cofferdam-core/src/adapter.rs) — describes the existing oxc-based TS/JS pipeline. language() returns Language::TypeScript; file_globs() returns ["*.ts", "*.tsx", "*.js", "*.jsx", "*.cjs", "*.mjs", "*.mts", "*.cts"], matching Language::from_path's extension list; schema_namespace() returns "ts".
• RustAdapter (crates/cofferdam-rust/src/lib.rs) — describes the existing tree-sitter-based Rust pipeline. language() returns Language::Rust; file_globs() returns ["*.rs"]; schema_namespace() returns "rust".
• HtmlAdapter (crates/cofferdam-html/src/lib.rs) — describes the tree-sitter-based HTML pipeline behind cofferdam verify --dist. language() returns Language::Html; file_globs() returns ["*.html", "*.htm"]; schema_namespace() returns "html".

Both impls are additive metadata: they describe existing behaviour without changing it. Neither cofferdam-engine's discovery loop nor its parse/ dispatch loop routes through these impls yet — file-extension dispatch still happens via Language::from_path exactly as before.

Not yet built

The following are explicitly deferred, not implied by this change:

• Runtime forbidden-surface enforcement. No mechanism stops an adapter from reading rule config or emitting Issues directly, and there is no deliberate-violation fixture test.
• [adapters] registration in cofferdam.toml. Adapters are not configurable or discoverable via config today.
• Adapter identity scheme for non-byte-offset findings. Tracked separately; not addressed here.
• An adapter for a language with a different fact model — SQL migrations, say. HTML made three, but all three are file-and-span languages; nothing yet proves the shape holds for a domain that isn't.
• Engine rewiring. cofferdam-engine's discovery and parse/dispatch loops are untouched; they do not consult Adapter impls at runtime.