菜单
Capture learnings, edge cases, and pitfalls from completed tasks and maintain Gotchas sections in Agent Skills for Memory Persistence. Use when finishing a task, resolving a bug, completing a review, or when an agent made a mistake that should be prevented in future sessions.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Formulates an AI for Science research plan. Proposes how to integrate AI into the researcher's domain through web research and literature survey, and generates a concrete 研究計画書 (research plan document). Use when user doesn't know how to apply AI to their 研究テーマ (research theme), wants to create a 研究プラン (research plan), or wants to learn about AI for Science latest trends (最新動向).
Support post-award research management and reporting for SPReAD. Guide grant application procedures, interim progress review (3 months), final outcome report (6 months), accounting report preparation, budget reallocation between expense categories, research plan change approval requests, publication acknowledgment templates, and research data management implementation. Use when you want to know post-award procedures, write an interim report, create a final report, change budget allocations, reallocate between expense categories, change research plans, or learn how to write publication acknowledgments.
Draft the SPReAD proposal (research plan document). Integrate the research plan, Azure architecture design, and cost estimate to generate a proposal aligned with the 公募要領 requirements. Use when writing the proposal, finalizing the research plan document, or preparing submission documents.
Estimate Azure costs based on architecture design. Generate monthly and full research period (~180 days) cost estimates and produce cost breakdowns required for SPReAD budget applications. Use when estimating Azure costs, planning budgets, or filling in expense sections of the proposal.
Designs a research infrastructure architecture on Microsoft Azure based on the research plan. Generates a comprehensive design document including GPU clusters, ML pipelines, data storage, and network configuration. References Microsoft Learn MCP for optimal configurations. Use when user wants to design Azure構成 (Azure architecture), build GPUクラスタ (GPU cluster), or design MLパイプライン (ML pipeline).
Generate step-by-step experiment procedure documents for running AI research on deployed Azure infrastructure. Covers environment setup, data preparation, training execution, inference, result evaluation, and checkpoint/reproducibility management. Use when you want to create an experiment procedure guide, run experiments on Azure ML, execute GPU training jobs, set up experiment environments, or manage reproducibility. 実験手順書を作りたい、Azure上で実験を実行したい、学習ジョブを走らせたい、 実験環境をセットアップしたい、再現性を管理したい場合に使用。
| name | gotchas-curator |
| description | Capture learnings, edge cases, and pitfalls from completed tasks and maintain Gotchas sections in Agent Skills for Memory Persistence. Use when finishing a task, resolving a bug, completing a review, or when an agent made a mistake that should be prevented in future sessions. |
| metadata | {"author":"coreclaw","version":"1.0"} |
タスク完了時やエージェントのミス発生時に学びを収集し、関連スキルの Gotchas セクションに反映するスキル。Harness 7軸の Memory Persistence を実践する。
対象イベントを特定:
学びを構造化:
Gotcha エントリを生成:
関連スキルの Gotchas セクションに追記
重複チェック:
## Gotchas
# ✅ 具体的で実行可能
- `users` テーブルはソフトデリートを使用。クエリには必ず
`WHERE deleted_at IS NULL` を含めること
- JWT トークンの `exp` クレームなしのトークンを許容してはならない。
必ず有効期限を検証すること
# ❌ 汎用的で価値が低い
- セキュリティに注意すること
- テストを忘れないこと
.github/learnings/ — スキルに紐づかない汎用的な学びの場合Gotchas はファイルに書き出されるため、セッションのコンパクション後も保持される。 以下のものはコンパクションで失われるため、重要なものは必ずファイルに書き出すこと:
| ✅ 保持される | ❌ 失われる |
|---|---|
| SKILL.md の Gotchas | セッション中の推論・分析 |
| ファイルに書き出した学び | 過去に読んだファイル内容 |
| Git にコミットした変更 | ツール呼び出し履歴 |