Struct LintEnv 

pub struct LintEnv {
    pub known_tools: HashSet<String>,
    pub known_models: Vec<(String, String)>,
    pub available_providers: Option<HashSet<String>>,
    pub read_paths: Option<Result<GrantReport, String>>,
}

Facts about the machine the blueprint will run on, which the manifest alone cannot supply.

Every field is “unknown” when empty/None, and an unknown field skips its check entirely rather than guessing. A linter that cannot see the installed MCP servers must not claim their tools do not exist.

Fields

known_tools: HashSet<String>

Every tool name a manifest may legally write: canonical built-ins, their aliases, the sub-agent tools, this agent’s own tools/*.rhai, and any MCP tools already resolved. Empty skips the unknown-tool check.

known_models: Vec<(String, String)>

(provider, model) rows for providers whose catalog is closed enough to check against. A provider with no row here is not checked at all, which is what keeps open catalogs (Ollama, OpenRouter, script providers) from producing noise.

available_providers: Option<HashSet<String>>

The providers the blueprint names that this install can actually reach, as answered by ProviderRegistry::has. None means nobody asked, so the check is skipped. Resolution lives with the caller because script providers are loaded on demand and cannot be enumerated up front.

read_paths: Option<Result<GrantReport, String>>

Which of the blueprint’s [read_paths] this install’s config grants. None means nobody asked (the daemon’s offline lint), in which case the check only says that a declaration needs granting. Some(Err(..)) is a grant list of the user’s own that will not compile.

Implementations

impl LintEnv

pub fn offline(agent_dir: &Path) -> Self

Everything that can be known without touching the user’s config: the built-in tools (aliases included), the sub-agent tools, the script tools in agent_dir/tools and the global tools directory, and the model catalogs this build ships.

This is what the daemon lints against at spawn. It deliberately leaves available_providers unset: the daemon already fails a spawn outright when no listed provider is registered, so re-deriving that here would cost a registry build per agent to say something the spawn will say louder a moment later.

pub fn with_providers(self, blueprint: &Blueprint, config: &Config) -> Self

Add the answer to “can this install reach the providers the blueprint names”, asked of the same registry the runtime resolves stages against so a script provider counts exactly when it would really load.

pub fn with_read_paths( self, blueprint: &Blueprint, config: &Config, workdir: &Path, ) -> Self

Add the answer to “does this install’s config grant what the blueprint declares under [read_paths]”, per entry.

Separate from Self::with_providers because it needs a workdir: relative entries resolve against the one a run would use, which for a command run outside a run is the directory it was invoked from.

Trait Implementations

impl Clone for LintEnv

fn clone(&self) -> LintEnv

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for LintEnv

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for LintEnv

fn default() -> LintEnv

Returns the “default value” for a type.