Have your skill file tested

• Cases fixed before the run. Around 20 tasks the skill is meant for, including the traps it claims to handle, each with a check written in code before any model sees it.
• Two models, with and without the skill. Claude Sonnet and Claude Haiku answer every case once with your skill loaded and once from the bare request, with tools switched off.
• Every answer read. The checks count first; then we read the answers, passing ones too, because a check can miss a wrong answer or fail a right one.

• The count for each model, with and without the skill, and what the difference is made of.
• The cases it fails, quoted, and what the skill could say to fix them.
• What the test did not measure.

The reports on our own skills show the format: see, for example, the tests of translate-app-text. A skill that does not help is a result too, and we say so.

• Write to hello@aiskills402.com with the subject "Skill test", the SKILL.md, and a few examples of the tasks it should handle.
• We reply with the test plan: the cases we would run and what would count as right.
• You pay only if you agree to the plan: $29 in USDC on Base to 0x8e37…36f4. Then we run the test and send the report.

Your skill and its report stay private: we do not publish them, list them or reuse your text. The file is run only through Anthropic's Claude models. The working copies of your skill, the cases and the report are deleted within 30 days of sending the report; the email thread stays in our mailbox until you ask us to delete it. See the privacy notice.