Tool Picker

"It depends" is the default answer to AI tool questions and it's the wrong one. People asking which tool to use have limited time and want a starting point that isn't a hedge. This skill gives one — opinionated, grounded in actual capability differences, not brand loyalty. The user can disagree with the recommendation, but they get an answer they can act on instead of a paragraph that ends in "try them and see."

The tool landscape is broken down by what each is genuinely best at and what it's weak at: Claude for long context, reasoning, and nuanced writing; Claude Code for in-codebase work with file access; ChatGPT for image generation, voice, and broad plugins; Perplexity for cited research and recent information; Cursor for IDE-integrated editing with visual diffs; Copilot for line-by-line completion; Midjourney, DALL-E, and Flux for images at different tradeoffs; Gemini for Google Workspace and video. Pricing tiers move constantly — recommendations are based on capability, with a note to check current pricing rather than baking it in.

The chat-vs-API distinction is treated as a first-class question, because most users don't realise it. If the task is interactive and one-off, the chat interface is correct. If it's batch processing, automation, or product integration, the API is correct — and "Claude" is not the same recommendation as "Claude API." Volume jobs in a chat window are a common failure mode the skill catches early.

It's honest about the tradeoff of "one tool for everything": one tool will cover about 70% of needs well; the other 30% will be noticeably weaker than a specialist. The skill names the right primary tool by use case (coding, research, writing, visual, Google ecosystem) instead of pretending one tool is universal. It also handles pushback gracefully — when a user has strong tool loyalty, the skill doesn't argue; it gives one concrete example where a different tool measurably outperforms for the stated task, then respects their choice if they stick. The best tool is the one they'll actually use.