Struct AnalyzeCtx 

pub struct AnalyzeCtx<'a> {
    pub provider: &'a dyn SourceProvider,
    pub kinds: &'a [StepKindSpec],
    pub kind_to_engine: &'a BTreeMap<String, String>,
    pub env: &'a BTreeMap<String, String>,
    pub config_vars: &'a BTreeMap<String, String>,
    pub run_id: &'a str,
    pub fragments: &'a FragmentCorpus,
}

Everything analyze_suite needs, injected at the IO edge (sans-IO core).

Fields

provider: &'a dyn SourceProvider

The source of feature and pack bytes (the IO edge lives behind it).

kinds: &'a [StepKindSpec]

Registered engine step kinds (drives pack validation and artifact probes).

kind_to_engine: &'a BTreeMap<String, String>

Step-kind prefix → engine id, the lowering routing table.

env: &'a BTreeMap<String, String>

Injected environment snapshot (${env:…}).

config_vars: &'a BTreeMap<String, String>

Injected proef.toml config scope (${url:…} / ${vars:…}), with the active [env.<name>] already deep-merged in.

run_id: &'a str

Injected run identifier (${run:id}).

fragments: &'a FragmentCorpus

The fragment corpus (ADR-0018), read by the caller.

Injected rather than built here, for the same reason every other input is: core performs no IO. It also fixes what building it internally cost — a fresh corpus means a fresh scan memo, so the LSP re-read and re-hurl-parsed the whole corpus on every request: each completion popup, each go-to-definition, each debounce tick. The caller holds one and rebuilds it only when a fragment file actually changes.