Struct SkillTool 

pub struct SkillTool { /* private fields */ }

Load a discovered skill package’s body on demand.

Holds the discovered INDEX (names, descriptions, manifest paths) that crate::tools::ToolRegistry::from_config resolved for this config — never a body. The set is fixed for the tool’s lifetime, the same way the prompt’s index section is: a skill installed mid-session appears after the next agent construction, exactly as it would in the harnesses this transcribes.

Implementations

impl SkillTool

pub fn new(skills: Vec<LoopSkill>) -> Self

Build the tool over an already-discovered skill set. Loads bodies with no shell injection — see Self::with_shell.

pub fn with_shell(self, shell: ShellInjection) -> Self

BP-5: the same tool, loading bodies under shell’s authorization — what crate::tools::ToolRegistry::from_config builds, so the tool door and the /name door expand a body identically.

pub fn skills(&self) -> &[LoopSkill]

The skills this tool can load.

pub fn find(&self, name: &str) -> Option<&LoopSkill>

Resolve an invocation name — crate::skills::find_skill, the same resolver /name and $slug go through.

Trait Implementations

impl Tool for SkillTool

fn name(&self) -> &str

Stable, unique tool name (what the model calls).

fn description(&self) -> &str

The built-in description. May be overridden via crate::Config.

fn parameters(&self) -> Value

JSON Schema describing the tool’s input object.

fn execute<'life0, 'life1, 'async_trait>( &'life0 self, args: Value, _ctx: &'life1 ToolContext, ) -> Pin<Box<dyn Future<Output = Result<String>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Run the tool. Returns text to feed back to the model.

fn structured_output(&self) -> bool

Whether a successful textual result is also a complete JSON value that protocol adapters should expose as structured output. Text remains the model-facing representation, preserving compatibility.