ai-tooling

"Analyze, evaluate, and optimize prompts for LLMs -- improve clarity, reduce token usage, add structure, and test variations" argument-hint: "<prompt text or file path> [--model claude|gpt|gemini] [--optimize-for clarity|tokens|reliability] [--compare]". TRIGGER WHEN: the user requires assistance with tasks related to this domain. DO NOT TRIGGER WHEN: the task is outside the specific scope of this component.

Expert prompt engineer for designing, optimizing, and managing prompts for LLMs. TRIGGER WHEN: writing system prompts, designing agent instructions, or optimizing prompt performance for reliability and token efficiency. DO NOT TRIGGER WHEN: the user is asking for general coding tasks unrelated to prompt engineering.

"DO NOT. TRIGGER WHEN: ALWAYS at conversation start and before every task. DO NOT TRIGGER WHEN: dispatched as a subagent."

You MUST use this before ANY creative or implementation work -- creating features, building components, adding functionality, modifying behavior, or any non-trivial task. Explores user intent, requirements and design before implementation. A conversation that starts as a question but progresses into "which approach should we take?" or "here's the architecture" IS a brainstorming session and MUST invoke this skill before any implementation begins. DO NOT. TRIGGER WHEN: (1) user says 'brainstorm', 'design', 'build', 'create', 'add', 'implement', 'develop', 'make', or gives any task involving new code; (2) the conversation has evolved into architectural discussion -- comparing approaches, evaluating trade-offs, drawing architecture diagrams, making technology choices, or discussing how components connect -- even if no explicit trigger word was used. A conversation that starts as a question but progresses into "which approach should we take?" or "here's the architecture" IS a brainstorming session and MUST invoke this skill before any implementation begins. DO NOT TRIGGER WHEN: user has a spec/plan ready, is fixing a specific bug, or doing a one-line change.

Build apps with the Claude Agent SDK (formerly Claude Code SDK). DO NOT. TRIGGER WHEN: code references claude-agent-sdk, user says "agent sdk", "build an agent", "programmatic claude", "claude code sdk", "sidecar", "run claude programmatically". DO NOT TRIGGER WHEN: user is using the Claude API client SDK (`anthropic`/`@anthropic-ai/sdk`) for direct chat completions, or doing general programming unrelated to agent orchestration.

"Also DO NOT. TRIGGER WHEN: user says 'execute the plan', 'run the plan', 'implement the plan', 'start implementing'. DO NOT TRIGGER WHEN: user wants to write a plan (use writing-plans), wants to brainstorm (use brainstorming), or is doing ad-hoc implementation."

Also use after brainstorming when the task involves 3+ files or multiple implementation steps. A conversation that evolved through brainstorming into a confirmed design MUST invoke this skill before writing any code, even if the user never explicitly said "write a plan". DO NOT. TRIGGER WHEN: (1) user says 'write a plan', 'create a plan', 'implementation plan', 'plan this', 'break this into tasks'; (2) the conversation has produced a design, spec, or set of decisions and is naturally transitioning toward implementation -- e.g., the user approved an approach, confirmed architecture choices, or said "let's do it" / "go ahead" / "proceed". A conversation that evolved through brainstorming into a confirmed design MUST invoke this skill before writing any code, even if the user never explicitly said "write a plan". DO NOT TRIGGER WHEN: user wants to brainstorm first (use brainstorming), wants to execute an existing plan (use executing-plans), or is doing a simple one-file change.