54 lines
1.5 KiB
Markdown
54 lines
1.5 KiB
Markdown
---
|
|
name: improve-prompt
|
|
description: Diagnose and improve an LLM prompt — fix ambiguity, add constraints, specify output format, add examples and edge case handling.
|
|
disable-model-invocation: true
|
|
argument-hint: "[prompt-text-or-file]"
|
|
context: fork
|
|
agent: prompt-engineer
|
|
---
|
|
|
|
# Improve Prompt
|
|
|
|
Diagnose and improve the provided prompt.
|
|
|
|
## Input
|
|
|
|
$ARGUMENTS
|
|
|
|
## Steps
|
|
|
|
1. **Diagnose the current prompt:**
|
|
- Ambiguity: vague instructions, unclear scope
|
|
- Missing output format: no schema or structure specified
|
|
- Weak constraints: "try to", "avoid if possible"
|
|
- No examples: complex tasks without few-shot
|
|
- Missing edge cases: no error/fallback handling
|
|
- No safety rules: missing refusal/deferral instructions
|
|
- Token bloat: redundant or filler text
|
|
|
|
2. **Improve following these principles:**
|
|
- Constraints before instructions (what NOT to do first)
|
|
- Explicit output schema with required fields and types
|
|
- 2-3 representative examples for complex tasks
|
|
- Edge case handling (empty input, malicious input, ambiguous request)
|
|
- Refusal rules for user-facing prompts
|
|
- Remove every sentence that doesn't change model behavior
|
|
|
|
3. **Verify via context7** — check target model capabilities and best practices
|
|
|
|
4. **Output:**
|
|
|
|
```markdown
|
|
## Diagnosis
|
|
- [Bullet list of issues found in the original prompt]
|
|
|
|
## Improved Prompt
|
|
[Clean, copy-ready prompt with clear sections]
|
|
|
|
## Changes Made
|
|
- [What changed and why — 3-5 items max]
|
|
|
|
## Usage Notes
|
|
- [Model, temperature, any caveats — only if non-obvious]
|
|
```
|