Struct SkillTriggersHook 

pub struct SkillTriggersHook { /* private fields */ }

On before_ingest, automatically activate relevant skills based on the user prompt. When a match is found, insert a L1 hint (e.g. “Detected skill X relevance; use the skill tool if needed”) before the prompt, rather than injecting the full skill body. This follows progressive disclosure: the model decides whether to actually load the skill.

Trigger conditions (any one triggers):

• keyword: any of the skill’s triggers.keywords is a case-insensitive substring of the prompt text.
• glob: any “path-like token” extracted from the prompt text matches one of the skill’s triggers.globs.

Skills with always trigger are already injected in full at session start, so they are skipped here to avoid duplicate hints.

Like SkillManifestHook, this hook holds a skill index and is registered via a closure that captures the index (see defect_cli::hooks).

Implementations

impl SkillTriggersHook

pub fn new(skills: Arc<BTreeMap<String, SkillEntry>>) -> Self

Constructs from the already-loaded skill index. The caller must not register this hook when skills is empty.

Trait Implementations

impl StepHandler for SkillTriggersHook

fn handle_step<'a>( &'a self, envelope: &'a Value, _ctx: HookCtx<'a>, ) -> BoxFuture<'a, Result<Option<Value>, HookError>>

In the before_ingest step, read the prompt text and, for each matched skill, prepend an L1 hint (a prepend_input verdict). Return None if no skill matches.