---
name: review-a-workflow
description: Review a real workflow by mapping people, steps, tools and pain points, then propose the smallest reversible improvement with human checkpoints and a measurable baseline.
---

# Review a workflow

Use this skill when someone wants to understand or improve a workflow involving real people, steps and tools. Adapt the review to the user's task; do not impose a large universal process.

## Start with what is known

Ask for, or extract from the supplied material:

- the current steps in their actual order;
- who performs, receives or approves each handoff;
- which tools, documents or systems are involved;
- the point where time, confusion, rework or risk appears;
- what can be observed now as a baseline.

Label missing facts as unknown. Separate a stated fact from an interpretation and from a suggestion. Do not infer savings, adoption, accuracy or a system change from a description alone.

## Make the smallest useful map

Represent each meaningful handoff as `person or role → action → artifact or decision → next handoff`. Mark where information is lost, duplicated, delayed or checked. Keep the map short enough that the owner can correct it.

For each pain point, test the options in this order:

1. **Eliminate:** remove a step, field or handoff that does not protect a real need.
2. **Simplify:** reduce choices, clarify the artifact or move a check closer to the decision.
3. **Automate:** suggest assistance only when the input and output are clear and a person can review the result.

Automation is a proposal, never an implied live change. Preserve a human checkpoint wherever an error could affect a person, deadline, money, access or an external message.

## Recommend a reversible next move

Choose one smallest change that can be tried without committing the whole team or replacing the current process. State:

- the exact step it changes;
- the owner and reviewer;
- the artifact produced;
- the checkpoint that can reject or correct it;
- the rollback path;
- one or two baseline measures to compare before and after.

Use measures the user can actually observe, such as elapsed time for a sample, number of handoffs, correction count or unanswered fields. Do not create a target or claim an improvement before it is measured.

## Return

Return a compact review with:

1. known facts and open questions;
2. current handoff map;
3. eliminate/simplify/automate options with their human checkpoint;
4. one reversible recommendation and rollback;
5. baseline and follow-up measures;
6. assumptions and limits.

Do not send messages, edit a production system, or claim that a workflow has changed. The user decides whether to run the suggested experiment.

See [EXAMPLE.md](EXAMPLE.md) for a fictional event-registration review.