Turn messy customer interview notes into a prioritized, testable experiment backlog.
Input: Raw customer interview notes (unstructured, contradictory, incomplete)
Output: A ranked markdown table of 5–10 experiments, each with:
- A clear, falsifiable hypothesis
- Measurable success signal
- ICE score for prioritization
The skill extracts pain points, clusters related signals, and converts each problem into exactly one testable experiment.
- Summarize interviews
- Brainstorm features
- Produce roadmaps
- Merge multiple problems into one experiment
This skill is designed to be model-agnostic and can be used anywhere.
Copy the folder into your skills directory. The core instructions live in SKILL.md.
Evaluated using Upskill:
- Sonnet is the interesting one: token usage drops while accuracy improves (!)
- The skill makes Sonnet more direct and concise while still meeting the rubric.
- The reduction in token usage suggests the skill constrains Sonnet’s tendency to over-elaborate, resulting in more direct and test-aligned outputs.
Recommended: Sonnet (best accuracy + lowest cost)
MIT