New open-source skill rewrites old Claude prompts using Anthropic's own model guides
A free tool called prompt-polish addresses a common problem: prompts written the old way can actually hurt on newer s. Many existing 'prompt improvers' take a short two-line request and blow it up into a 500-word manual full of assigned roles, section headings, micro-steps, phrases like 'think step by step' and 'be thorough,' and dozens of negative instructions. That approach clashes with what Anthropic's own model-specific guides recommend.
Fable 5 often performs better with less prescription, not more, and Anthropic itself warns that prompts or skills written for earlier models can be overly prescriptive and degrade output quality on newer ones. Opus 4.8 behaves the opposite way: it follows instructions very literally, so any scope the writer only implied rather than stated can turn into a bug. The same generic 'improvement' can therefore help one model while hurting another.
prompt-polish rewrites rough prompts by the official guide for whichever model you're targeting, rather than applying one universal template. It installs via 'npx skills add mfarzanansari/prompt-polish,' and the code is available on GitHub.
Key points
- Prompts written in the old verbose style can hurt on newer s
- Fable 5 tends to do better with less prescriptive instructions rather than more
- Opus 4.8 follows instructions very literally, so implied scope left unstated can become a bug
- prompt-polish rewrites prompts based on the official guide for the specific instead of one universal template
- Install with 'npx skills add mfarzanansari/prompt-polish'; the code is on GitHub