aiw-prompt-smithlisted
Install: claude install-skill philippe-ths/ai-coding-workflow
# Prompt Smith
Two jobs, one spine. Use this when **writing** a prompting artifact (system prompt, CLAUDE.md, AGENTS.md/AGENT.md, another skill, an agent prompt file, or any file an agent loads to learn what to do, whatever it is named) or **repairing** one after it produced a failure you noticed. Moving instructions from one file to another is repair: something about where they lived has failed.
The governing idea is **altitude**. Every instruction sits at some level of generality, and most prompt problems are altitude errors. Too low and you write a brittle rule for the one case in front of you, and this is what breeds bloat. Too high and you write something so vague it steers nothing. Aim for the altitude where a competent role or a named concept does the work.
This skill is deliberately small. If your edit makes an artifact longer without making it clearer, you are probably patching the instance, not fixing the class.
## Start here (both modes)
Before writing or editing anything, get four answers. If you are repairing, get a fifth.
1. **Behaviour**: what should the artifact make the model do?
2. **Failure it prevents**: what goes wrong without this? If nothing, you do not need the instruction.
3. **One example**: a concrete good-vs-bad pair, ideally an awkward case rather than an easy one.
4. **Layer**: who reads this line, in which file, with what else in their context at the time? Altitude is about the line; layer is about where it lives, and it is a decision, not