メニュー
Model selection guide for oa agents: haiku/sonnet/opus by task type. Use when choosing --model for oa run or oa pipeline. Activates for: which model, claude/haiku, claude/sonnet, claude/opus, model selection, --model.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/Impertio-Studio/Open-Agents --skill oa-prompting-model-tieringコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Bidirectioneel communicatiesysteem tussen meta-orchestrator en agents. Auto-loads bij agent spawning, inbox monitoring, statusupdates ontvangen, of feedback van agents. Triggert op: oa send, oa inbox, oa watch-inbox, agent done, blocked agent, feedback loop, terugkoppeling.
Applies OpenAEC Foundation brand to artifacts. WORKSPACE OVERRIDE — overrides global Impertio brand. Triggers on: branding, huisstijl, corporate identity, visual design, OpenAEC style.
Detects newly created or recently modified skills in ~/.claude/skills/ and auto-generates matching agent templates in the Open-Agents library. Use when a new skill has been created, when skills have been updated, or when the user says 'generate template for this skill', 'scan new skills', 'update agent library from skills', or 'watch skill changes'.
| name | oa-prompting-model-tiering |
| description | Model selection guide for oa agents: haiku/sonnet/opus by task type. Use when choosing --model for oa run or oa pipeline. Activates for: which model, claude/haiku, claude/sonnet, claude/opus, model selection, --model. |
| user-invocable | false |
claude/sonnet ID format for --model, never claude-sonnet-4-6 — because oa run uses short model aliases, not Anthropic API IDs.--model explicitly — because the default may change across oa-cli versions, producing unpredictable behavior.| Task Type | Model | Rationale |
|---|---|---|
| Scanning, listing, parsing, formatting | claude/haiku | Fast, cheap; structured output only |
| Writing, coding, implementation | claude/sonnet | DEFAULT — balanced quality/speed |
| Architecture, deep reasoning, complex orchestration | claude/opus | Maximum depth; use sparingly |
| QA, validation, review | claude/sonnet | Needs judgment, not just speed |
| Research (broad) | claude/sonnet | Good breadth coverage |
| Research (deep analysis) | claude/opus | When depth > speed |
| Batch processing (repetitive) | claude/haiku | Cost efficiency at scale |
Task type?
├── Scanning / listing / formatting / parsing → claude/haiku
├── Writing / coding / implementation / review → claude/sonnet (DEFAULT)
└── Architecture / deep reasoning / orchestration planning → claude/opus
oa run "Implement the auth module" --name builder --model claude/sonnet --direct
oa run "Plan the architecture" --name planner --model claude/opus --direct
oa run "Convert CSV to JSON" --name converter --model claude/haiku --direct
oa run "Plan all implementation steps" --name planner --model claude/opus --direct
oa run "Implement step 1" --name builder-a --model claude/sonnet --direct
oa run "Convert file to JSON" --name converter --model claude/haiku --direct
--model claude — bare alias, version-unstable--model claude/sonnet — explicit tier--model anthropic/claude-sonnet-4-6 — Anthropic API ID, not oa-cli alias--model claude/sonnetclaude/opus for all tasks — expensive and slow for simple work