Prompt Fix

Install in one command

mkdir -p ~/.claude/skills/prompt-fix && curl -fsSL https://eliteaiadvantage.com/skills/prompt-fix/SKILL.md -o ~/.claude/skills/prompt-fix/SKILL.md

When a prompt isn't producing what you want, the failure mode is rewriting the whole thing and re-running. You change three things at once, the output gets slightly different, you can't tell which change helped, you change three more things. Prompt Fix forces single-variable ablations: change one thing per run, log results, compare.

The skill names the failure mode precisely (format / reasoning / consistency / tone), defines success concretely, proposes three single-variable ablations, runs each, and outputs a comparison table. No more vibing on prompt rewrites.

• →Forces precision on the failure mode — "outputs sometimes return markdown when JSON expected" beats "the prompt isn't great."
• →Each ablation changes one variable relative to baseline — so you can attribute improvement.
• →Runs each ablation 3x where possible to handle LLM non-determinism.
• →Outputs a structured comparison table, not a vibe.
• →Saves winning ablations as new versions instead of overwriting working prompts.

• •A prompt is misbehaving and your three rewrites haven't helped.
• •Output is right "most of the time" but inconsistent.
• •Migrating a prompt to a new model and behavior changed.
• •Tuning a prompt for production where you need predictable output.

Without this skill, prompt iteration is essentially gambling: change phrasing, run, eyeball, repeat. Prompt Fix imposes engineering discipline on the loop — one variable per change, multiple runs per ablation, comparison against explicit success criteria. The output is reproducible, not a vibe-driven rewrite.