| SKILL.md | name, description | allowed-tools, model, context, agent, hooks, user-invocable | | Agent .md | name, description | tools, disallowedTools, model, permissionMode, skills, hooks | | Command .md | (none) | description, allowed-tools, model, argument-hint | | plugin.json | name, description | author | | hooks.json | hooks object | description |
| High | opus | Architecture, debugging, multi-domain reasoning | | Medium | sonnet | Implementation, refactoring, moderate analysis | | Low | haiku | Searches, simple transforms, quick lookups | | Inherited | inherit | When parent context model is appropriate |
| Complex multi-step | plan | High complexity requiring human feedback before implementation | | Edge case handling | acceptEdits | Potential edge cases need user confirmation on edits | | Routine execution | bypassPermissions | Clear requirements, preplanned, or promise completion | | Standard workflow | default | Normal permission checking with prompts |