Prompts / heuristic_evaluation
Nielsen heuristic evaluation of a flow Free
Walks one task through a flow and records findings against Nielsen's ten heuristics with 0 to 4 severity, using the worksheet tool for signals. You do the judging.
Run it
- Claude Code:
/mcp__designers__heuristic_evaluation - Claude Desktop and claude.ai: plus menu, Add from Designers, Nielsen heuristic evaluation of a flow
- Any app: Use the heuristic_evaluation prompt from the Designers server on figma link, url, screenshots or a description of the flow.
Arguments
| Name | Meaning |
|---|---|
flow * | Figma link, URL, screenshots or a description of the flow |
task * | The one task to evaluate, in the user's words |
audience | Who is doing the task |
* required
What your assistant is told to do
The full instructions, so you can see exactly what you are asking for:
Do a heuristic evaluation. It is a human-style judgement method: you walk a task and record problems. The tool gives signals; the findings are yours.
Flow: <flow>
Task: <task>
Person doing it: <audience>
Method:
1. Read resource designers://reference/nielsen-heuristics for the heuristics, the walkthrough method and the severity scale.
2. Get the flow. If the input is a Figma link: parse it (figma.com/design/:fileKey/..., node-id=1-2 means nodeId "1:2"), call get_metadata to see the frames, get_screenshot for each frame you will judge, and get_design_context for layout, text and tokens. Read the picture before the code. If the Figma tools are not connected, say so and ask for screenshots or exported frames instead; do not guess at what is in the file. For a live URL, also call audit_page_accessibility and heuristic_worksheet with the url and task to collect signals.
3. Write the task as a list of steps a person takes. At each step note what they are trying to do, what they see, what they will do next.
4. For each problem write: step, what happens, heuristic number and name, effect on the person, severity 0 to 4 (rate after you have the full list, using frequency, difficulty to get past, and cost of failure), fix direction. Only record what would plausibly hurt a real person in this task. Do not log taste.
5. Run heuristic_worksheet with the flow's copy and the task to check you did not miss a heuristic that its signals point at; add findings only if you agree.
6. Output a table sorted by severity, then three sentences: the top three issues, one thing that works, and what a real evaluator would still need to test (states and devices you did not see). Say plainly that one evaluator finds roughly a third of problems and recommend two more independent passes.
If you have Pro, you can feed the findings to build_heuristic_report for a scored, exportable markdown report.