Compose semantics

Compose merges multiple skills into a single artifact suitable for injecting into an agent's system prompt.

Rules

• Version ranges follow semver.
• Conflicts resolved by: caller preference → trust tier → composite score → LLM judge.
• Every output section is tagged with provenance: <!-- from mobile/ios-publishing@2.3.1 -->.
• The sources[] array lists every skill that contributed to the merge.
• The conflicts[] array records when two skills covered the same heading + how it was resolved.

Conflict resolution strategies

• error — fail the request if any two skills overlap.
• trust-tier (default) — prefer the higher-tier skill's content.
• score — prefer whichever skill the registry's hybrid ranker scored higher.
• llm-judge — ask the judge to merge intelligently. Slower; better for nuanced overlaps.

Example

const pkg = await skills.compose({
  goal: 'publish iOS+Android app',
  constraints: { trust_floor: 'verified', resolve_conflicts: 'trust-tier' },
});

// pkg.artifact:
// <!-- from mobile/ios-publishing@0.1.0 -->
// # iOS
// ...
// <!-- from mobile/android-play-release@0.1.0 -->
// # Android
// ...

// pkg.sources:
// [
//   { id: 'mobile/ios-publishing', version: '0.1.0', trust_tier: 'verified' },
//   { id: 'mobile/android-play-release', version: '0.1.0', trust_tier: 'verified' }
// ]