AI Workflow Designer

Most people automate the wrong things. They either pick something trivial that saves two minutes a week, or they try to automate a complex judgement call that AI can't reliably do — and when it fails, they conclude AI doesn't work for their role. The sweet spot is in between: tasks that are repeatable, time-consuming, knowledge-heavy but not deep-judgement, and produce structured outputs. This skill exists to find that sweet spot in your specific work.

It runs a four-phase discovery — role mapping, opportunity scoring, workflow design, and a tool-specific recommendation — instead of dumping a generic list of "10 things to automate". Every candidate task gets scored on three axes: Time Saved (hours per week), Automation Feasibility (can AI actually do it well), and Risk (what happens if AI gets it wrong). The formula is explicit — Time × Feasibility ÷ Risk — so you can see why a boring 3-hours-saved task beats a flashy 15-minute one, and why an unsafe-if-wrong task drops below the line even if it's tempting.

For each high-priority task you get a full workflow design, not just a prompt: Trigger (what kicks it off), Inputs (what info Claude needs and where it lives), Process (what Claude actually does), Output (format and length), Review Point (where the human checks), Delivery (where it goes), and Setup Time (honest estimate — 10 minutes for a simple prompt, 1-2 hours for a multi-tool pipeline, half a day for anything needing MCP setup).

The skill is opinionated about what not to automate. For sensitive professions — clinical, legal, financial — it names the professional-judgement work as human-only and points at the admin surround instead. For users who are already well-optimised, it doesn't invent problems to solve; it asks about remaining friction and suggests sharpening existing workflows rather than adding new ones. The rule it enforces hardest is "start with one workflow, not ten" — because ten half-built workflows is the most common way these audits fail.