endorphin-ai/hasbrains-agent-kit / context-engineering/skills/prompt-expert
prompt-expert
FlaggedExpert prompt engineer that interviews users and builds high-quality prompts for any AI (Claude, Gemini, Copilot, ChatGPT, etc.). Use when a user wants to create a prompt, system prompt, or AI instructions from scratch — especially beginners who know what they want to achieve but don't know how to write it. Trigger on: "help me write a prompt", "make me a prompt for...", "I want AI to do X", "build me a system prompt", "how do I ask AI to...", "create instructions for...", or any time the user describes a goal they want an AI to accomplish. Always interviews the user before writing anything. Never completes the user's described task directly — always treats the request as a prompt engineering job.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 1.1 KB |
| SKILL.md | skill md | 9.2 KB |
| references/ai-tailoring.md | markdown | 5.5 KB |
| references/engineering-patterns.md | markdown | 5.3 KB |
| references/output-templates.md | markdown | 4.8 KB |
| references/safety-security.md | markdown | 4.8 KB |
| references/tester-protocol.md | markdown | 2.9 KB |
references/ai-tailoring.md
View on GitHub# AI-Specific Prompt Tailoring
Platform quirks, syntax preferences, and concrete examples for each major AI.
---
## Claude (Anthropic)
**Best for:** Complex reasoning, long documents, structured analysis, code, writing.
**What works well:**
- XML tags to separate sections — Claude parses these reliably
- `<thinking>` tag to force reasoning before answer
- Long, detailed system prompts — Claude handles them well
- Negative constraints ("never", "do not") — high compliance
**Syntax pattern:**
```
<role>
You are [ROLE]. [1-2 sentences describing behavior.]
</role>
<instructions>
TASK: [What to do]
RULES:
- ALWAYS [behavior]
- NEVER [behavior]
</instructions>
<format>
- Structure: [list / paragraph / JSON]
- Length: [short / medium / exact word count]
- Tone: [formal / casual]
</format>
<context>
[PLACEHOLDER: paste your content here]
</context>
```
**Tips:**
- Put the most important instruction last — Claude has recency bias
- Use `<thinking>` for math, logic, multi-step analysis
- Explicit "do not apologize", "do not add disclaimers" reduces hedging
---
## Gemini (Google)
**Best for:** Factual research, summarization, multimodal (text + image), Google Workspace integration.
**What works well:**
- Markdown headers (##, ###) to structure prompts
- Explicit numbered steps
- Clear "Output format:" section
- Concrete examples
**Syntax pattern:**
```
## Role
You are [ROLE]. [Purpose.]
## Task
[Imperative description of what to do.]
## Input
[PLACEHOLDER: user will provide X]
## Steps
1. [Step 1]
2. [Step 2]
3. [Step 3]
## Output format
- Structure: [format]
- Length: [length]
- Tone: [tone]
## Constraints
- Never [rule 1]
- Always [rule 2]
```
**Tips:**
- Gemini responds well to "think step by step" for reasoning tasks
- For Google Workspace: reference specific apps ("in Google Docs format")
- Shorter prompts often work better than very long ones
---
## ChatGPT / GPT-4 (OpenAI)
**Best for:** Conversational tasks, coding, creative writing, general-purpose assistance.
**What works well:**
- Strong role assignment in system prompt
- Few-shot examples (2-3 input/output pairs) — very effective
- Numbered constraint lists
- "Respond only with..." to control output format
**System prompt pattern:**
```
You are [ROLE], a [description].
Your job: [core task in 1 sentence].
Rules:
1. Always [rule]
2. Never [rule]
3. [rule]
Output format: [explicit format description]
```
**User prompt pattern:**
```
[TASK INSTRUCTION]
Input: [PLACEHOLDER]
Respond only with [format]. Do not include [unwanted element].
```
**Few-shot example pattern:**
```
Examples of correct output:
Input: [example 1]
Output: [expected output 1]
Input: [example 2]
Output: [expected output 2]
Now do the same for:
Input: [PLACEHOLDER]
```
**Tips:**
- Few-shot examples are the single highest-ROI technique for GPT models
- "Respond only with X" is very effective for format control
- Temperature guidance: mention "be precise" for factual, "be creative" for generative
---
## Copilot — GitHub (Microsoft)
**Best for:** Code generation, code review, refactoring, documentation, test writing.
**What works well:**
- Explicit language and framework context
- Acceptance criteria style output spec
- Reference to file structure or existing patterns
- Concrete input (paste the code directly in the prompt)
**Syntax pattern:**
```
TASK: [What to do with the code]
LANGUAGE: [Python / TypeScript / Go / etc.]
FRAMEWORK: [React / FastAPI / etc.]
STYLE: [PEP8 / Airbnb / project conventions]
INPUT:
[PLACEHOLDER: paste code here]
REQUIREMENTS:
- [Requirement 1]
- [Requirement 2]
OUTPUT:
- Return: [what to produce]
- Format: [inline comments / full file / diff]
- Include: [tests / docstrings / types]
- Do NOT change: [what to preserve]
```
**Tips:**
- Always specify what NOT to change — prevents over-editing
- Include test requirements explicitly or it will skip them
- "Explain your changes" produces better-reasoned output
---
## Copilot — Microsoft 365 (Microsoft)
**Best for:** Email drafting, document summarization, meeting notes, Excel formulas, PowerPoint outlines.
**What works well:**
- Action-oriented, short prompts
- Reference to specific M365 app context
- Explicit output length ("3 bullet points", "1 paragraph")
- Tone specification for business context
**Syntax pattern:**
```
[ACTION VERB] [OBJECT] for [PURPOSE].
Context: [PLACEHOLDER: paste email / document / data]
Output:
- Format: [bullet list / email / table / summary]
- Length: [short / 3 bullets / 1 paragraph / under 100 words]
- Tone: [professional / friendly / executive]
Constraints:
- Do not include [unwanted element]
- Focus only on [specific aspect]
```
**Tips:**
- Keep prompts short — M365 Copilot has a limited effective context window
- Always specify tone — business context varies widely
- For emails: specify recipient role ("to a client", "to my team")
---
## Universal / AI-Agnostic
Use when the target AI is unknown or the prompt must work across multiple platforms.
**Rules:**
- Plain imperative English only — no platform-specific syntax
- No XML tags, no markdown headers (some platforms don't parse them)
- Numbered lists for steps and rules — universally readable
- Keep it short — optimize for the most constrained platform
**Pattern:**
```
You are [ROLE]. [1 sentence purpose.]
Your task: [What to do — imperative]
Input: [PLACEHOLDER]
Always:
1. [Required behavior]
2. [Required behavior]
Never:
1. [Prohibited behavior]
2. [Prohibited behavior]
Output: [format] of [length] in [tone] tone.
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |