Prompt Engineering Expert

### Quick judgment A strong, specialized meta-skill for users who actively design prompts, system messages, and agent instructions. It appears well-scoped to prompt engineering best practices, optimization, and evaluation, and will be most valuable for building complex agents, reusable templates, or robust workflows rather than for casual one-off questions. ### Strengths - **Deep focus on prompt craft**: Covers prompt writing, custom instructions, optimization, advanced patterns (few-shot, chain-of-thought, role-based, XML-style structure), and anti-pattern identification. - **Evaluation & iteration**: Emphasis on test cases, success criteria, and debugging prompt performance is particularly useful for teams shipping prompt-based features. - **Context & multimodal awareness**: Explicit handling of context window management, token efficiency, and multimodal prompting (vision, embeddings, file-based prompts) increases practicality for real-world systems. ### Risks and limitations - **Redundancy with built-in capabilities**: Modern models already provide decent prompt help; this skill’s value depends on how well its internal guidance actually exceeds default behavior. Without code or tests, that’s inferred rather than verified. - **Complexity & verbosity**: Advanced techniques (chain-of-thought, heavy few-shot, complex XML schemas) can make prompts long, expensive, and harder to maintain if overused. - **Possible conflicts with higher-level instructions**: Aggressive system-prompt design or role framing could clash with platform-level policies or other system prompts if not managed carefully. - **No direct guarantee of empirical performance**: The description promises methodology but doesn’t prove that its patterns consistently outperform simpler prompts across domains. ### Recommended scenarios Use this skill when: - You’re **designing or refactoring system prompts/custom instructions** for agents, tools, or skills and want structured guidance. - You need to **debug inconsistent or brittle behavior** from an AI system and want a principled way to iterate and test prompts. - You’re building **reusable prompt templates** or prompt libraries for a product or internal tooling. - You’re **teaching prompt engineering** or formalizing standards for a team (e.g., code review-like processes for prompts). - You want to **optimize context usage and structure** for long or multimodal queries, or integrate embeddings/files into prompt flows. It is less critical for simple, ad-hoc Q&A, but well-suited as a scaffolding layer for anyone treating prompts as a first-class engineering artifact.