# ste-prompt-fixer
An agent skill that fixes sloppy prompts. It rewrites them in controlled-English style (ASD-STE100, ~80% strict) and picks the right output format: plain text, diagram, or single-file HTML.
Inspired by Andrej Karpathy's X post (Oct 2, 2026) on understanding LLM outputs.
Topics: `llm` `prompt-engineering` `claude-skills` `asd-ste100` `agents` `agent-skills` `mcp` `technical-writing` `turkish-llm`
## Why
Models write long, padded answers by default. Controlled languages fix that: short sentences, fixed vocabulary, no ambiguity. This skill wraps that idea so any agent that reads `SKILL.md` (Claude Code, Codex, Cursor, Amp, Copilot) can apply it.
Bonus: `lint/ste-lint.md` turns the same rules into a linter spec, and `tests/fixtures.md` provides seed prompts for regression tests.
## Install
Claude Code: copy `SKILL.md` to `~/.claude/skills/prompt-fixer/SKILL.md`.
Cursor: paste the body into `.cursor/rules/prompt-fixer.mdc`.
Any agent following the agentskills.io format: place the folder in its skills path.
## Structure
```
SKILL.md core flow: intent -> rewrite -> format
profiles/ style packs, one standard per file
ste80.md ASD-STE100, 80% strict (default)
plain-english.md public plain-language style
globish.md 1500-word ceiling, non-native readers
cte-style.md procedure-first service manual style
turkish.md Turkish humanizer + STE hybrid
conventions/ context packs
agents-md.md align prompts with repo AGENTS.md rules
llms-txt.md suggest /llms.txt for site-targeted prompts
lint/
ste-lint.md linter-style rules + regex checks for LLM output
tests/
fixtures.md bad prompt -> fixed prompt seed set (goal: 50)
examples/
before-after.md one fixed prompt per profile
```
The model reads a profile file only when that profile is selected. Keeps context small.
## Format ladder
```mermaid
flowchart LR
A[Raw prompt] --> B[Detect intent]
B --> C[Apply profile: ste80 default]
C --> D{Pick format}
D -->|facts| E[Plain text]
D -->|relations, flows| F[Mermaid / SVG]
D -->|dense, explorable| G[Single-file HTML]
```
## Example
Input:
> you are a senior dev please explain microservices vs monolith thanks urgent
Output (ste80):
```
Describe microservices vs monolith.
Audience: a team choosing a stack for a new project.
Give one table, max 6 rows: latency, deploy speed, debugging, scaling, team size, cost.
End with a 3-sentence verdict: when to choose each.
Max 300 words.
```
## ste-lint quick check
```bash
grep -EIn "(worth noting|delve|tapestry|ever-evolving|in today's fast-paced world|game-changer)" docs/*.md && exit 1
```
## Credits
- Andrej Karpathy for the ASD-STE100 prompting idea
- ASD for the STE100 specification
- Caterpillar, IBM, GM for decades of controlled-language work
## License
MIT