Expose `disableModelInvocation` on `session.skills_loaded` skill entries

session.skills_loaded イベントスキーマとスキルごとのエントリ型から始め、Rust の SkillsLoadedSkill と TypeScript の session.skills_loaded データ型を含めて確認します。次に、関連する Skill と ServerSkill の discovery 形状、および producer に既存の model-invocation フラグを調べます。完了条件は、一覧にあるすべての形状が disableModelInvocation を公開し、producer が既存のフラグを使ってそれを設定し、古い producer のデフォルト値は false のままであることです。

Summary

The session.skills_loaded event reports each skill's enabled and userInvocable flags, but not its model-invocation flag (the disable-model-invocation frontmatter gate). Please add disableModelInvocation to each skill entry so consumers can tell whether the model may auto-invoke a skill — not just whether the user can invoke it as a slash command.

Why

userInvocable (user slash-command availability) and model-invocability are two independent axes:

• userInvocable: true/false — can the user invoke it as a slash command.
• disableModelInvocation: true/false — may the model auto-invoke it.

A skill can be any combination (user-only, model-only, both). Consumers today only receive userInvocable, so they can't distinguish a model-auto-invoked skill from a user-only one. This matters for:

• Rendering an accurate command palette / skill list (which entries are slash commands vs. silently model-driven).
• Projecting skills into surfaces that have a native model-invocation concept (for example, AHP's SkillCustomization already has a disableModelInvocation field, which currently can't be populated from this event).

The flag is already part of the skill model — it's used internally to decide whether a skill is visible to the model — so this is a surfacing change, not new behavior.

Current shape

session.skills_loaded skill entries today (camelCase wire):

{
  "name": "…",
  "description": "…",
  "source": "project | inherited | personal-copilot | personal-agents | plugin | custom | builtin",
  "path": "…",            // optional
  "argumentHint": "…",    // optional
  "enabled": true,
  "userInvocable": true
}

Proposed change

Add disableModelInvocation: boolean to each entry:

{
  "name": "…",
  "description": "…",
  "source": "…",
  "path": "…",
  "argumentHint": "…",
  "enabled": true,
  "userInvocable": true,
  "disableModelInvocation": false   // NEW
}

Surfaces to touch:

• The session.skills_loaded event schema (the per-skill entry shape).
• Generated types in every language target (e.g. the Rust SkillsLoadedSkill and the TypeScript session.skills_loaded data type).
• For consistency, mirror the field onto the related skill API types that also omit it today (the Skill / ServerSkill discovery shapes).
• Populate it at the producer from the skill model's existing model-invocation flag.

Backwards compatibility

Additive boolean; existing consumers are unaffected. A sensible default for older producers is false (model invocation allowed), matching the default semantics of the disable-model-invocation frontmatter flag.