Builds the Agent Skills catalog, loading tool, and reserved tool context for a Jido.AI.Agent when the agent instance initializes.

Discovery is explicit because scanning a project loads instructions from its filesystem. Passing true trusts the standard project and user skill roots; passing a list trusts only those roots. Keyword options expose the discovery bounds and a custom trust predicate. Keyword options do not trust any root unless :trust is explicitly set.

Hosts may also pass runtime %Jido.AI.Skill.Spec{} values with source: nil and inline bodies:

agent_skills: [
  specs: runtime_specs,
  resource_provider: &MyApp.SkillResources.handle/2,
  resource_policy: [max_text_bytes: 131_072]
]

Runtime specs are validated, preserved without filesystem discovery or filesystem roots, and added to the same scoped catalog as discovered skills. Runtime specs take precedence over discovered skills with the same name; shadowed discovered entries are reported in diagnostics. Duplicate runtime names are rejected.

The catalog contains metadata-only discovered specs. Full files are read and strictly validated only after the model selects a filesystem skill. Runtime specs keep their inline body and metadata exactly as supplied.

Enabled agents receive both load_skill and load_skill_resource. The optional :resource_policy sets listing and text-loading limits for both actions.