gitea-actions
172 lines
7.5 KiB
Markdown
172 lines
7.5 KiB
Markdown
---
|
||
name: prompt-optimizer
|
||
description:
|
||
"Help users rewrite and improve AI/LLM prompts by adding specificity, context,
|
||
and constraints. Trigger this skill whenever users ask to improve, rewrite,
|
||
optimize, or refine prompts for AI models. Focus on making prompts clearer,
|
||
more specific, and more likely to produce better AI results. Present
|
||
suggestions interactively so users can choose which improvements to apply."
|
||
---
|
||
|
||
# Prompt Optimizer
|
||
|
||
A beginner-friendly skill for improving AI/LLM prompts to get better results.
|
||
|
||
## What This Skill Does
|
||
|
||
This skill helps you rewrite prompts to work better with AI models like Claude.
|
||
Instead of just giving you a rewritten prompt, it shows you specific improvement
|
||
suggestions that you can choose to apply or skip.
|
||
|
||
## Key Improvements
|
||
|
||
When optimizing a prompt, focus on three main areas:
|
||
|
||
### 1. **Specificity** — Making the Request Clear
|
||
|
||
Good prompts are specific about what you want. Vague prompts get vague results.
|
||
|
||
**Example improvements:**
|
||
|
||
- Add details about format: "Give me a bullet list of 5 items" instead of "tell
|
||
me about X"
|
||
- Be clear about length: "Write 200 words" instead of "Write something short"
|
||
- Define who the audience is: "Explain this for a 10-year-old" or "Use technical
|
||
language"
|
||
|
||
### 2. **Context** — Giving the AI Background Information
|
||
|
||
More context helps the AI make better decisions.
|
||
|
||
**Example improvements:**
|
||
|
||
- Explain the goal: "I'm writing a resume, so focus on professional language"
|
||
- Share constraints: "We only have $500 budget" or "It needs to work on mobile"
|
||
- Provide background: "I already know Python but not JavaScript"
|
||
|
||
### 3. **Constraints** — Setting Boundaries
|
||
|
||
Constraints prevent unwanted outputs.
|
||
|
||
**Example improvements:**
|
||
|
||
- Set length limits: "Keep it under 100 words"
|
||
- Specify format: "Use JSON format" or "Write as a numbered list"
|
||
- Define tone: "Be casual and friendly, not formal"
|
||
- Say what NOT to include: "Don't use technical jargon"
|
||
|
||
## How to Use This Skill
|
||
|
||
1. **Share your prompt** — Give me the original prompt you want to improve
|
||
2. **Review suggestions** — I'll show you specific improvements in each area
|
||
3. **Choose what you like** — Pick which suggestions to apply
|
||
4. **Get the final version** — I'll rewrite your prompt with your chosen
|
||
improvements
|
||
|
||
## Interactive Selection Process
|
||
|
||
When you use this skill, you'll see:
|
||
|
||
- **Original prompt** — Your starting point
|
||
- **Improvement suggestions** — Specific changes grouped by category
|
||
(Specificity, Context, Constraints)
|
||
- **Preview examples** — What each change would look like with the improvement
|
||
applied
|
||
- **Your choices** — You pick which suggestions help most (you can apply all,
|
||
some, or none) Then you get a rewritten prompt combining all your choices.
|
||
|
||
### Example Interaction
|
||
|
||
**Original:** "Write me a blog post"
|
||
|
||
**Suggestions I might offer:**
|
||
|
||
- **Specificity**: Add a topic (e.g., "about sustainable living")
|
||
- **Context**: Explain your goal (e.g., "to build authority on my website")
|
||
- **Constraints**: Set a word count (e.g., "800-1000 words") **Your choice:** "I
|
||
want all three — add topic, goal, and word count"
|
||
|
||
**Final rewritten prompt:** "Write an 800-1000 word blog post about sustainable
|
||
living for my website. The goal is to establish my authority on eco-friendly
|
||
practices. Target an audience of people interested in reducing their carbon
|
||
footprint. Include 3-4 practical tips they can implement immediately, and end
|
||
with a call-to-action encouraging them to sign up for my newsletter."
|
||
|
||
## Tips for Best Results
|
||
|
||
- **Start simple** — Even small improvements help
|
||
- **Focus on your goal** — What outcome do you want?
|
||
- **Add one constraint at a time** — Too many rules can be confusing
|
||
- **Test and iterate** — Try the new prompt and see if results improve
|
||
|
||
## What Makes a Good Prompt
|
||
|
||
A prompt becomes "good" when:
|
||
|
||
- The AI understands exactly what you want ✓
|
||
- You've given enough context to explain why ✓
|
||
- You've set boundaries to prevent bad outputs ✓
|
||
- Someone else could read it and understand your intent ✓
|
||
|
||
---
|
||
|
||
## Anti-Pattern Detection
|
||
|
||
Before suggesting improvements, scan the original prompt for these common
|
||
mistakes and report them with a warning:
|
||
|
||
| Anti-Pattern | Description | Example |
|
||
| ---------------------- | ----------------------------------------------------- | ------------------------------------------- |
|
||
| Too generic | No clear subject, action, or goal | "Tell me about AI" |
|
||
| Ambiguous pronouns | "it", "that", "this" with no clear referent | "Fix it so it works better" |
|
||
| Internal contradiction | Two requirements that cancel each other out | "Be concise but cover everything in detail" |
|
||
| Missing context | Requests an action without explaining why or for whom | "Rewrite this paragraph" |
|
||
|
||
**Output format:** Before showing improvement suggestions, print a "Detected
|
||
Issues" block listing any anti-patterns found. If none are found, skip this
|
||
block silently. Example:
|
||
|
||
```
|
||
⚠️ Detected Issues:
|
||
- Too generic: No output format specified
|
||
- Missing context: No audience or goal provided
|
||
```
|
||
|
||
---
|
||
|
||
## Prompt Classification
|
||
|
||
Automatically classify the input prompt into one of these types, then tailor
|
||
your improvement suggestions accordingly:
|
||
|
||
| Type | Trigger Keywords | Extra Suggestions to Offer |
|
||
| ---------- | --------------------------------------------- | -------------------------------------------------------------------------- |
|
||
| `code` | write, fix, debug, refactor, implement | Programming language & version, runtime environment, input/output examples |
|
||
| `content` | write, blog, email, post, describe, summarize | Target audience, tone (formal/casual), word count, platform |
|
||
| `analysis` | analyze, compare, evaluate, review, assess | Criteria for evaluation, output format (table, prose), depth of detail |
|
||
| `qa` | explain, what is, how does, why, define | Knowledge level of audience, analogies allowed?, length of answer |
|
||
| `task` | do, create, set up, build, generate, automate | Step-by-step vs one-shot, tools/permissions available, success criteria |
|
||
|
||
Show the detected type at the top of your response:
|
||
`📌 Prompt type detected: [type]`
|
||
|
||
---
|
||
|
||
## Chain-of-Thought Mode
|
||
|
||
Offer "Chain-of-Thought Enhancement" as an optional improvement when showing
|
||
suggestions. When the user selects it, add reasoning instructions to the end of
|
||
the optimized prompt based on type:
|
||
|
||
| Prompt Type | Chain-of-Thought Addition |
|
||
| --------------- | -------------------------------------------------------------------------- |
|
||
| General | `"Think step by step before answering."` |
|
||
| Analysis | `"Explain your reasoning for each point before giving a conclusion."` |
|
||
| Code | `"Outline your approach and data structures before writing any code."` |
|
||
| Decision-making | `"List pros and cons, then state your recommendation with justification."` |
|
||
|
||
**Why it works:** Chain-of-Thought forces the AI to externalize its reasoning
|
||
process. This reduces hallucination (the AI catches its own errors
|
||
mid-reasoning), makes outputs easier to verify, and produces more structured
|
||
answers. Studies show CoT improves accuracy on complex tasks by 20–40%.
|