Agent Skills: llm-feature-design
アプリに組み込む LLM 機能 (チャット・要約・分類・RAG など) のシステムプロンプトと周辺構造を、AI ベンダー公式・OWASP・査読論文で裏付けたベストプラクティスに沿って設計する。回答範囲の制限、非信頼テキストの扱いとプロンプトインジェクション対策、ハルシネーション抑制、ガードレールの多層化、評価セットの作り方を references/ から参照する。「LLM 機能を実装したい」「システムプロンプトを設計して」「AI が関係ない質問に答えてしまう」「プロンプトインジェクション対策」「AI の回答範囲を絞りたい」「RAG の回答を文書に基づかせたい」などで起動。Claude Code 自身への作業依頼プロンプトの整形は対象外 (prompt-refine を使う)。
ID: skanehira/dotfiles/llm-feature-design
884
Install this agent skill to your local
Skill Files
The files that make up llm-feature-design.
More agent skills
- goplacesQuery Google Places for text search, place details, resolve, reviews, or scriptable JSON via goplaces.
- audit-asc-prAudit App-Store-Connect-CLI pull requests end to end and fix concrete defects when authorized. Use for PR review, value and blast-radius assessment, issue verification, or requested fixes before merge.
- wox-generate-smoke-test-caseAnalyze Wox feature code and plugin-exposed settings, recommend high-value native UI smoke scenarios and user flows, then generate focused Go UI smoke test cases using the automation driver and shared-process harness. Use when the user asks to identify smoke coverage for a Wox feature or plugin, or to design, add, write, or generate a smoke case under wox.core/test/smoke.
- query-layerUse when editing TanStack Query integration: createQuery, createMutation, resultQueryOptions, resultMutationOptions, defineQuery, defineMutation, defineKeys, shared cache identity, mutation lifecycle, or service-to-TanStack adapters.
- dimensional-analysisAnnotates codebases with dimensional analysis comments documenting units, dimensions, and decimal scaling. Use when someone asks to annotate units in a codebase, perform a dimensional analysis, or find vulnerabilities in a DeFi protocol, offchain code, or other blockchain-related codebase with arithmetic. Prevents dimensional mismatches and catches formula bugs early.
- prior-auth-packet-builderBuild a concise prior authorization packet from local case files and payer policy docs.