Prompt Tuner

$2.99Official

Use when a prompt underperforms and needs systematic diagnosis, single-variable experiments, and A/B evidence instead of guess-and-check rewrites.

agent-infrastructureprompt-engineeringab-testingdebuggingevalsoptimization· v1· by SkillingMain
63
Usefulness score
850
Installs
No ratings yet
today
Last updated

Model requirements

Capability tier

advanced

Min context window

33k tokens

Recommended models
Claude Sonnet 4GPT-4o-miniGemini 2.5 FlashLlama 3.3 70B (self-hosted)Qwen 2.5 72B (self-hosted)Mistral Large

Preview

When to use

Use this skill when a prompt (system prompt, instruction block, tool description, or template) is underperforming and the user wants it fixed with evidence rather than vibes: wrong formats, ignored instructions, hallucinated fields, inconsistent tone, misfired tool calls. Also use it when the user proposes a prompt rewrite and wants to know if it is actually better.

Do not use it when the failure is caused by missing information or a missing tool — no wording fixes an impossible task.

Inputs to gather

  1. The exact current prompt, verbatim, including any template variables and where runtime content is injected.
  2. Failure examples: at least 5 real input/output pa

🔒 Buy this skill once ($2.99) to unlock the full playbook and install it anywhere — or get everything with Pro.